Chapter 1

1. In case of an accident, there is a high chance of getting lost. The transportation cost is very high eachtime. However, if the infrastructure is set once, it will be very easy to use it repeatedly. Time forwireless transmission is negligible as signals travel at the speed of light.

2. Advantages of bursty data communication

(a) Pulses are made very narrow, so multipaths are resolvable

(b) The transmission device needs to be switched on for less time.

Disadvantages

(a) Bandwidth required is very high

(b) Peak transmit power can be very high.

3. Pb = 10−12

12γ = 10−12

γ = 1012

2 = 5× 1011 (very high)

4. Geo: 35,786 Km above earth ⇒ RTT = 2×35786×103

c = 0.2386sMeo: 8,000- 20,000 Km above earth ⇒ RTT = 2×8000×103

c = 0.0533s

Leo: 500- 2,000 Km above earth ⇒ RTT = 2×500×103

c = 0.0033sOnly Leo satellites as delay = 3.3ms < 30ms

5.

6. optimum no. of data user = doptimum no. of voice user = vThree different cases:Case 1: d=0, v=6⇒ revenue = 60.80.2 = 0.96

Case 2: d=1, v=3revenue = [prob. of having one data user]×(revenue of having one data user)+ [prob. of having two data user]×(revenue of having two data user)+ [prob. of having one voice user]×(revenue of having one voice user)+ [prob. of having two voice user]×(revenue of having two voice user)+ [prob. of having three or more voice user]×(revenue in this case)

⇒ 0.52

(21

)× $1 + 0.52 × $1 +

(61

)0.8× 0.25 × $0.2 +

(62

)0.82 × 0.24 × $0.4+

[1−

(61

)0.8× 0.25 × $0.2−

(62

)0.82 × 0.24 × $0.4

]× $0.6

⇒ $1.35

Case 3: d=2, v=0revenue =2× 0.5 = $1So the best case is case 2, which is to allocate 60kHz to data and 60kHz to voice.

7.

8. 1. Hand-off becomes a big problem.2. Inter-cell interference is very high and should be mitigated to get reasonable SINR.3. Infrastructure cost is another problem.

9. Smaller the reuse distance, larger the number of users who can use the same system resource and socapacity (data rate per unit bandwidth) increases.

10. (a) 100 cells, 100 users/cell ⇒ 10,000 users

(b) 100 users/cell ⇒ 2500 cells required100km2

Area/cell = 2500cells ⇒ Areacell = .04km2

(c) From Rappaport or iteration of formula, we get that 100 channelscell ⇒ 89 channels

cell @Pb = .02Each subscriber generates 1

30 of an Erlang of traffic.Thus, each cell can support 30× 89 = 2670 subscribersMacrocell: 2670× 100 ⇒ 267, 000 subscribersMicrocell: 6,675,000 subscribers

(d) Macrocell: $50 MMicrocell: $1.25 B

(e) Macrocell: $13.35 M/month ⇒ 3.75 months approx 4 months to recoopMicrocell: $333.75 M/month ⇒ 3.75 months approx 4 months to recoop

11. One CDPD line : 19.2Kbpsaverage Wimax ∼ 40Mbps∴ number of CDPD lines ∼ 2× 103

Chapter 2

1.

Pr = Pt

[√Glλ

4πd

]2

λ = c/fc = 0.06

10−3 = Pt

[λ

4π10

]2

⇒ Pt = 4.39KW

10−3 = Pt

[λ

4π100

]2

⇒ Pt = 438.65KW

Attenuation is very high for high frequencies

2. d= 100mht = 10mhr = 2mdelay spread = τ = x+x′−l

c = 1.33×

3. ∆φ = 2π(x′+x−l)λ

x′ + x− l =√

(ht + hr)2 + d2 −√

(ht − hr)2 + d2

= d

√(ht + hr

d

)2

+ 1−√(

ht − hr

d

)2

+ 1

d À ht, hr,we need to keep only first order terms

∼ d

1

2

√(ht + hr

d

)2

+ 1

−

1

2

√(ht − hr

d

)2

+ 1

=2(ht + hr)

d

∆φ ∼ 2π

λ

2(ht + hr)d

4. Signal nulls occur when ∆φ = (2n + 1)π

2π(x′ + x− l)λ

= (2n + 1)π

2π

λ

[√(ht + hr)2 + d2 −

√(ht − hr)2 + d2

]= π(2n + 1)

√(ht + hr)2 + d2 −

√(ht − hr)2 + d2 =

λ

2(2n + 1)

Let m = (2n + 1)√

(ht + hr)2 + d2 = mλ

2+

√(ht − hr)2 + d2

square both sides

(ht + hr)2 + d2 = m2 λ2

4+ (ht − hr)2 + d2 + mλ

√(ht − hr)2 + d2

x = (ht + hr)2, y = (ht − hr)2, x− y = 4hthr

x = m2 λ2

4+ y + mλ

√y + d2

⇒ d =

√[1

mλ

(x−m2

λ2

4− y

)]2

− y

d =

√(4hthr

(2n + 1)λ− (2n + 1)λ

4

)2

− (ht − hr)2 , n ∈ Z

5. ht = 20mhr = 3mfc = 2GHz λ = c

fc= 0.15

dc = 4hthrλ = 1600m = 1.6Km

This is a good radius for suburban cell radius as user density is low so cells can be kept fairly large.Also, shadowing is less due to fewer obstacles.

6. Think of the building as a plane in R3

The length of the normal to the building from the top of Tx antenna = ht

The length of the normal to the building from the top of Rx antenna = hr

In this situation the 2 ray model is same as that analyzed in the book.

7. h(t) = α1δ(t− τ) + α2δ(t− (τ + 0.22µs))Gr = Gl = 1ht = hr = 8mfc = 900MHz, λ = c/fc = 1/3R = −1

delay spread =x + x′ − l

c= 0.022× 10−6s

⇒2√

82 +(

d2

)2 − d

c= 0.022× 10−6s

⇒ d = 16.1m

∴ τ =d

c= 53.67ns

α1 =(

λ

4π

√Gl

l

)2

= 2.71× 10−6

α2 =(

λ

4π

√RGr

x + x′

)2

= 1.37× 10−6

8. A program to plot the figures is shown below. The power versus distance curves and a plot of thephase difference between the two paths is shown on the following page. From the plots it can be seenthat as Gr (gain of reflected path) is decreased, the asymptotic behavior of Pr tends toward d−2 fromd−4, which makes sense since the effect of reflected path is reduced and it is more like having only aLOS path. Also the variation of power before and around dc is reduced because the strength of thereflected path decreases as Gr decreases. Also note that the the received power actually increases withdistance up to some point. This is because for very small distances (i.e. d = 1), the reflected path isapproximately two times the LOS path, making the phase difference very small. Since R = -1, thiscauses the two paths to nearly cancel each other out. When the phase difference becomes 180 degrees,the first local maxima is achieved. Additionally, the lengths of both paths are initially dominated bythe difference between the antenna heights (which is 35 meters). Thus, the powers of both paths areroughly constant for small values of d, and the dominant factor is the phase difference between thepaths.

clear all;close all;ht=50;hr=15;f=900e6;c=3e8;lambda=c/f;GR=[1,.316,.1,.01];Gl=1;R=-1;counter=1;figure(1);d=[1:1:100000];l=(d.^2+(ht-hr)^2).^.5;r=(d.^2+(ht+hr)^2).^.5;phd=2*pi/lambda*(r-1);dc=4*ht*hr/lambda;dnew=[dc:1:100000];

for counter = 1:1:4,Gr=GR(counter);Vec=Gl./l+R*Gr./r.*exp(phd*sqrt(-1));Pr=(lambda/4/pi)^2*(abs(Vec)).^2;subplot(2,2,counter);plot(10*log10(d),10*log10(Pr)-10*log10(Pr(1)));hold on;plot(10*log10(dnew),-20*log10(dnew));plot(10*log10(dnew),-40*log10(dnew));

endhold off

0 20 40 60−150

−100

−50

0

50

10 *

log1

0(P

r)

10 * log10(d)

Gr = 1

0 20 40 60−150

−100

−50

0

50

10 *

log1

0(P

r)

10 * log10(d)

Gr = .316

0 20 40 60−150

−100

−50

0

50

10 *

log1

0(P

r)

10 * log10(d)

Gr = .1

0 20 40 60−150

−100

−50

0

50

10 *

log1

0(P

r)

10 * log10(d)

Gr = .01

0 20 40 600

100

200

300

400

Pha

se (

deg)

10 * log10(d)

Figure 1: Problem 8

9. As indicated in the text, the power fall off with distance for the 10-ray model is d−2 for relatively largedistances

10. The delay spread is dictated by the ray reaching last d =√

(500/6)2 + 102 = 83.93mTotal distance = 6d = 503.59mτ0 = 503.59/c = 1.68µsL.O.S ray d = 500mτ0 = 500/c = 1.67µs∴ delay spread = 0.01µs

11. fc = 900MHzλ = 1/3mG = 1 radar cross section 20dBm2 = 10 log1 0σ ⇒ σ = 100d=1 , s = s′ =

√(0.5d)2 + (0.5d)2 = d

√0.5 =

√0.5

Path loss due to scattering

Pr

Pt=

[λ√

Gσ

(4π)3/2ss′

]2

= 0.0224 = −16.498dB

Path loss due to reflection (using 2 ray model)

Pr

Pt=

(R√

G

s + s′

)2 (λ

4π

)2

= 3.52× 10−4 = −34.54dB

d = 10 Pscattering = −56.5dB Preflection = −54.54dBd = 100 Pscattering = −96.5dB Preflection = −74.54dBd = 1000 Pscattering = −136.5dB Preflection = −94.54dB

Notice that scattered rays over long distances result in tremendous path loss

12.

Pr = PtK

(d0

d

)γ

→ simplified

Pr = Pt

(√Gl

4π

)2 (λ

d

)2

→ free space

∴ when K =(√

Gl4π

)2and d0 = λ

The two models are equal.

13. Pnoise = −160dBmfc = 1GHz, d0 = 1m, K = (λ/4πd0)2 = 5.7× 10−4, λ = 0.3, γ = 4We want SNRrecd = 20dB = 100∵ Noise power is 10−19

P = PtK

(d0

d

)γ

10−17 = 10K

(0.3d

)4

d ≤ 260.7m

14. d = distance between cells with reused freqp = transmit power of all the mobiles

(S

I

)

uplink

≥ 20dB

(a) Min. S/I will result when main user is at A and Interferers are at BdA = distance between A and base station #1 =

√2km dB = distance between B and base station

#1 =√

2km(

S

I

)

min

=P

[Gλ

4πdA

]2

2P[

Gλ4πdB

]2 =d2

B

2d2A

=(dmin − 1)2

4= 100

⇒ dmin − 1 = 20km ⇒ dmin = 21km since integer number of cells should be accommodated indistance d ⇒ dmin = 22km

(b)

Pγ

Pu= k

[d0

d

]γ

⇒(

S

I

)

min

=Pk

[d0dA

]γ

2Pk[

d0dB

]γ =

12

[dB

dA

]γ

=12

[dmin − 1√

2

]γ

=12

[dmin − 1√

2

]3

= 100

⇒ dmin = 9.27 ⇒ with the same argument ⇒ dmin = 10km

(c)(

S

I

)

min

=k

[d0dA

]γ

A

2k[

d0dB

]γ

B

=(dmin − 1)4

0.04= 100

⇒ dmin = 2.41km ⇒ with the same argument dmin = 4km

15. fc = 900MHz, ht = 20m,hr = 5m, d = 100m

Large urban city PLlargecity = 353.52dBsmall urban city PLsmallcity = 325.99dBsuburb PLsuburb = 207.8769dBrural area PLruralarea/countryside = 70.9278dB

As seen , path loss is higher in the presence of multiple reflectors, diffractors and scatterers

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3−60

−50

−40

−30

−20

−10

0Plot for Gr = 0 dB

y = −15x+8

y = −35x+56

Figure 2: Problem 16

16. Piecewise linear model for 2-path model. See Fig 2

17. Pr = Pt − PL(d)−∑3i FAFi −

∑2j PAFj

FAF =(5,10,6), PAF =(3.4,3.4)

PL(d)K(

d0

d

)γ

0

= 10−8 = −8dB

−110 = Pt − 80− 5− 10− 6− 3.4− 3.4

⇒ Pt = −2.2dBm

18. (a) PrPt

dB = 10 log10 K − 10r log1 0 dd0

using least squares we get10 log10 K = −29.42dBγ = 4

(b) PL(2Km) = 10 log10 K − 10r log10 d = −161.76dB

(c) Receiver power can be assumed to be Gaussian with variance σ2ψdB

X ∼ N(0, σ2ψdB)

Prob(X < −10) = Prob

(X

σψdB<−10σψdB

)= 6.512× 10−4

19. Assume free space path loss parametersfc = 900MHz → λ = 1/3mσψdB = 6SNRrecd = 15dBPt = 1W

g = 3dBPnoise = −40dBm ⇒ Precvd = −55dBSuppose we choose a cell of radius d

µ(d) = Precvd(due to path loss alone)

= Pt

(√Glλ

4πd

)2

=1.4× 1−−3

d2

µdB = 10 log1 0(µ(d))P (Precd(d) > −55) = 0.9

P

(Precd(d)− µdB

σψdB>−55− µdB

6

)= 0.9

⇒ −55− µdB

6= −1.282

⇒ µdB = −47.308⇒ µ(d) = 1.86× 10−5

⇒ d = 8.68m

20. MATLAB CODE

Xc = 20;ss = .01;y = wgn(1,200*(1/ss));for i = 1:length(y)

x(i) = y(i);for j = 1:i

x(i) = x(i)+exp(-(i-j)/Xc)*y(j);end

end

21. Outage Prob. = Prob. [received powerdB 6 TpdB]Tp = 10dB

(a)

outageprob. = 1−Q

(Tp− µψ

σψ

)= 1−Q

(−58

)= Q

(58

)= 26%

(b) σψ = 4dB, outage prob < 1% ⇒

Q

(Tp− µψ

σψ

)> 99% ⇒ Tp− µψ

σψ< −2.33 ⇒

µψ ≥ 19.32dB

(c)

σψ = 12dB,Tp− µψ

σψ< −6.99 ⇒ µψ ≥ 37.8dB

0 20 40 60 80 100 120 140 160 180 200−60

−50

−40

−30

−20

−10

0

10

20White Noise ProcessFiltered Shadowing Processing

Figure 3: Problem 20

(d) For mitigating the effect of shadowing, we can use macroscopic diversity. The idea in macroscopicdiversity is to send the message from different base stations to achieve uncorrelated shadowing.In this way the probability of power outage will be less because both base stations are unlikely toexperience an outage at the same time, if they are uncorrelated.

22.

C =2

R2

R∫

r=0

rQ(a + b ln

r

R

)dr

To perform integration by parts, we let du = rdr and v = Q(a + b ln r

R

). Then u = 1

2r2 and

dv =∂

∂rQ

(a + b ln

r

R

)=

∂

∂xQ(x)|x=a+b ln(r/R)

∂

∂r

(a + b ln

r

R

)=

−1√2π

exp(−k2/2)b

rdr. (1)

where k = a + b ln rR . Then we get

C =2

R2

[12r2Q

(a + b ln

r

R

)]R

r=0

+2

R2

R∫

r=0

12r2 1√

2πexp(−k2/2)

b

rdr (2)

= Q(a) +1

R2

R∫

r=0

r2 1√2π

exp(−k2/2)b

rdr (3)

= Q(a) +1

R2

R∫

r=0

1√2π

R2 exp(

2(k − a)b

)exp(−k2/2)

b

rdr (4)

= Q(a) +

a∫

k=−∞

1√2π

exp(−k2

2+

2k

b− 2a

b

)dk (5)

= Q(a) + exp(−2a

b+

2b2

) a∫

k=−∞

1√2π

exp(−1

2(k − 2

b)2

)dk (6)

= Q(a) + exp(

2− 2ab

b2

)[1−Q

(a− 2

b

1

)](7)

= Q(a) + exp(

2− 2ab

b2

)Q

(2− ab

b

)(8)

(9)

Since Q(−x) = 1−Q(x).

23. γ = 3d0 = 1k = 0dBσ = 4dBR = 100mPt = 80mWPmin = −100dBm = −130dB

Pγ(R) = PtK

(d0

d

)γ

= 80× 10−9 = −70.97dB

a =Pmin − Pγ(R)

σ= 14.7575

b =10γ log10 e

σψdB= 3.2572

c = Q(a) + e2−2ab

b2 Q

(2− 2ab

b

)' 1

24. γ = 6σ = 8

Pγ(R) = 20 + Pmin

a = −20/8 = −2.5

b =10× 6× log10 e

8= 20.3871

c = 0.9938

25.

γ/σψdB2 4 6

4 0.7728 0.8587 0.89778 0.6786 0.7728 0.825512 0.6302 0.7170 0.7728

Since Pr(r) ≥ Pmin for all r ≤ R, the probability of non-outage is proportional to Q(−1

σ

), and thus

decreases as a function of σ. Therefore, C decreases as a function of σ. Since the average power at theboundary of the cell is fixed, C increases with γ, because it forces higher transmit power, hence morereceived power at r < R. Due to these forces, we have minimum coverage when γ = 2 and σ = 12. Bya similar argument, we have maximum coverage when γ = 6 and σ = 4. The same can also be seenfrom this figure:

2

3

4

5

6

4

6

8

10

120.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

γσψ

dB

Cov

erag

e

Figure 4: Problem 25

The value of coverage for middle point of typical values i.e. γ = 4 and σ = 8 can be seen from thetable or the figure to be 0.7728.

Chapter 3

1. d = vtr + r′ = d + 2h2

dEquivalent low-pass channel impulse response is given by

c(τ, t) = α0(t)e−jφ0(t)δ(τ − τ0(t)) + α1(t)e−jφ1(t)δ(τ − τ1(t))

α0(t) = λ√

Gl4πd with d = vt

φ0(t) = 2πfcτ0(t)− φD0

τ0(t) = d/cφD0 =

∫t 2πfD0(t)dt

fD0(t) = vλ cos θ0(t)

θ0(t) = 0 ∀tα1(t) = λR

√Gl

4π(r+r′) = λR√

Gl

4π(d+ 2h2

d)

with d = vt

φ1(t) = 2πfcτ1(t)− φD1

τ1(t) = (r + r′)/c = (d + 2h2

d )/cφD1 =

∫t 2πfD1(t)dt

fD1(t) = vλ cos θ1(t)

θ1(t) = π − arctan hd/2 ∀t

2. For the 2 ray model:

τ0 =l

c

τ1 =x + x′

c

∴ delay spread(Tm) =x + x′ − l

c=

√(ht + hr)2 + d2 −

√(ht − hr)2 + d2

c

when d À (ht + hr)

Tm =1c

2hthr

d

ht = 10m, hr = 4m, d = 100m

∴ Tm = 2.67× 10−9s

3. Delay for LOS component = τ0 = 23 nsDelay for First Multipath component = τ1 = 48 nsDelay for Second Multipath component = τ2 = 67 ns

τc = Delay for the multipath component to which the demodulator synchronizes.

Tm = maxm

τm − τc

So, when τc = τ0, Tm = 44 ns. When τc = τ1, Tm = 19 ns.

4. fc = 109Hzτn,min = 10

3×108 s

∴ min fcτn = 1010

3×108 = 33 À 1

5. Use CDF strategy.

Fz(z)= P [x2+y2 ≤ z2] =∫ ∫

x2+y2≤z2

12πσ2

e−(x2+y)2

2σ2 dxdy =

2π∫

0

z∫

0

12πσ2

e−r2

2σ2rdrdθ = 1− e

−z2

2σ2 (z ≥ 0)

dfz(z)dz

=z

σ2e−

z2

2σ2 → Rayleigh

For Power:

Fz2(z)=P [Z ≤ √z] = 1− e

−z2

2σ2

fz(z)=1

2σ2e−z

2σ2→ Exponential

6. For Rayleigh fading channelPr(Pr < P0) = 1− e−P0/2σ2

2σ2 = −80dBm, P0 = −95dBm, Pr(Pr < P0) = 0.0311

2σ2 = −80dBm, P0 = −90dBm, Pr(Pr < P0) = 0.0952

7. For Rayleigh fading channelPoutage = 1− e−P0/2σ2

0.01 = 1− e−P0/Pr

∴ Pr = −60dBm

8. 2σ2 = -80dBm = 10−11

Target Power P0 = -80 dBm = 10−11

Avg. power in LOS component = s2 = -80dBm = 10−11

Pr[z2 ≤ 10−11] = Pr[z ≤ 10−5

√10

]

Let z0 = 10−5√10

=∫ z0

0

z

σ2e

−−(z2+s2)

2σ2

I0

( zs

σ2

)dz, z ≥ 0

= 0.3457

To evaluate this, we use Matlab and I0(x) = besseli(0,x). Sample Code is given:

clear P0 = 1e-11; s2 = 1e-11; sigma2 = (1e-11)/2; z0 =sqrt(1e-11); ss = z0/1e7; z = [0:ss:z0]; pdf =(z/sigma2).*exp(-(z.^2+s2)/(2*sigma2)).*besseli(0,z.*(sqrt(s2)/sigma2));int_pr = sum(pdf)*ss;

9. CDF of Ricean distribution isFRicean

Z (z) =∫ z

0pRicean

Z (z)

where

pRiceanZ (z) =

2z(K + 1)Pr

exp[−K − (K + 1)z2

Pr

]I0

(2z

√K(K + 1)

Pr

), z ≥ 0

For the Nakagami-m approximation to Ricean distribution, we set the Nakagami m parameter to be(K + 1)2/(2K + 1). CDF of Nakagami-m distribution is

FNakagami-mZ (z) =

∫ z

0pNakagami-m

Z (z)

where

pNakagami-mZ (z) =

2mmz2m−1

Γ(m)Prmexp

[−mz2

Pr

], z ≥ 0, m ≥ 0.5

We need to plot the two CDF curves for K = 1,5,10 and Pr =1 (we can choose any value for Pr as itis the same for both the distributions and our aim is to compare them). Sample code is given:

z = [0:0.01:3]; K = 10; m = (K+1)^2/(2*K+1); Pr = 1; pdfR =((2*z*(K+1))/Pr).*exp(-K-((K+1).*(z.^2))/Pr).*besseli(0,(2*sqrt((K*(K+1))/Pr))*z);pdfN = ((2*m^m*z.^(2*m-1))/(gamma(m)*Pr^m)).*exp(-(m/Pr)*z.^2);for i = 1:length(z)

cdfR(i) = 0.01*sum(pdfR(1:i));cdfN(i) = 0.01*sum(pdfN(1:i));

end plot(z,cdfR); hold on plot(z,cdfN,’b--’); figure;plot(z,pdfR); hold on plot(z,pdfN,’b--’);

As seen from the curves, the Nakagami-m approximation becomes better as K increases. Also, for afixed value of K and x, prob(γ<x) for x large is always greater for the Ricean distribution. This is seenfrom the tail behavior of the two distributions in their pdf, where the tail of Nakagami-distribution isalways above the Ricean distribution.

10. (a) W = average received powerZi = Shadowing over link iPri = Received power in dBW, which is Gaussian with mean W, variance σ2

(b)

Poutage = P [Pr,1 < T ∩ Pr,2 < T ] = P [Pr,1 < T ]P [Pr,2 < T ] since Z1, Z2 independent

=[Q

(W − T

σ

)]2

=[Q

(4σ

)]2

(c)

Pout =

∞∫

−∞P [Pr,1 ≤ T , Pr,2 < T |Y = y] fy (y) dy

Pr,1|Y = y ~ N(W + by,a2σ2

), and [Pr,1|Y = y] ⊥ [Pr,2|Y = y]

Poutage=

∞∫

−∞

[Q

(W + by − T

aσ

)]2 1√2πσ

e−y2

2σ2 dy

0 0.5 1 1.5 2 2.5 30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

RiceanNakagami −m

0 0.5 1 1.5 2 2.5 30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1K = 5

RiceanNakagami−m

0 0.5 1 1.5 2 2.5 30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1K = 10

RiceanNakagami−m

0 0.5 1 1.5 2 2.5 30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1K = 1

RiceanNakagami − m

0 0.5 1 1.5 2 2.5 30

0.5

1

1.5K = 5

RiceanNakagami−m

0 0.5 1 1.5 2 2.5 30

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2K = 10

RiceanNakagami−m

3.5 4 4.5 5 5.5 60

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2x 10

−31

Tail Behavior

K = 10

RiceanNakagami−m

Figure 1: The CDF and PDF for K = 1, 5, 10 and the Tail Behavior

let yσ= u

=

∞∫

−∞

1√2π

[Q

(W − T + buσ

aσ

)]2

e−u2

2 du =

∞∫

−∞

1√2π

[Q

(4+ byσ

aσ

)]2

e−y2

2dy

(d) Let a = b = 1√2

, σ = 8, 4 = 5. With independent fading we get

Pout=[Q

(58

)]2

= 0.0708.

With correlated fading we get Pout= 0.1316.Conclusion : Independent shadowing is prefarable for diversity.

11. There are many acceptable techniques for this problem. Sample code for both the stochastic tech-nique(preferred) and the Jake’s technique are included.

Jakes: Summing of appropriate sine waves

%Jake’s Methodclose all; clear all;%choose N=30N=30; M=0.5*(N/2-1); Wn(M)=0; beta(M)=0;%We choose 1000 samples/secritemp(M,2001)=0; rqtemp(M,2001)=0; rialpha(1,2001)=0; fm=[1 10100]; Wm=2*pi*fm; for i=1:3

for n=1:1:Mfor t=0:0.001:2

%Wn(i)=Wm*cos(2*pi*i/N)Wn(n)=Wm(i)*cos(2*pi*n/N);%beta(i)=pi*i/Mbeta(n)=pi*n/M;%ritemp(i,20001)=2*cos(beta(i))*cos(Wn(i)*t)%rqtemp(i,20001)=2*sin(beta(i))*cos(Wn(i)*t)ritemp(n,1000*t+1)=2*cos(beta(n))*cos(Wn(n)*t);rqtemp(n,1000*t+1)=2*sin(beta(n))*cos(Wn(n)*t);%Because we choose alpha=0,we get sin(alpha)=0 and cos(alpha)=1%rialpha=(cos(Wm*t)/sqrt(2))*2*cos(alpha)=2*cos(Wm*t)/sqrt(2)%rqalpha=(cos(Wm*t)/sqrt(2))*2*sin(alpha)=0rialpha(1,1000*t+1)=2*cos(Wm(i)*t)/sqrt(2);

endend%summarize ritemp(i) and rialphari=sum(ritemp)+rialpha;%summarize rqtemp(i)rq=sum(rqtemp);%r=sqrt(ri^2+rq^2)r=sqrt(ri.^2+rq.^2);%find the envelope averagemean=sum(r)/2001;subplot(3,1,i);

time=0:0.001:2;%plot the figure and shift the envelope average to 0dBplot(time,(10*log10(r)-10*log10(mean)));titlename=[’fd = ’ int2str(fm(i)) ’ Hz’];title(titlename);xlabel(’time(second)’);ylabel(’Envelope(dB)’);

end

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2−10

−5

0

5fd = 1 Hz

Env

elop

e(dB

)

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2−20

−10

0

10fd = 10 Hz

Env

elop

e(dB

)

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2−20

−10

0

10fd = 100 Hz

Env

elop

e(dB

)

Figure 2: Problem 11

Stochastic: Usually two guassian R.V.’s are filtered by the Doppler Spectrum and summed. Can alsojust do a Rayleigh distribution with an adequate LPF, although the above technique is prefered.

function [Ts, z_dB] = rayleigh_fading(f_D, t, f_s)%% function [Ts, z_dB] = rayleigh_fading(f_D, t, f_s)% generates a Rayleigh fading signal for given Doppler frequency f_D,% during the time perios [0, t], with sampling frequency f_s >= 1000Hz.%% Input(s)% -- f_D : [Hz] [1x1 double] Doppler frequency% -- t : simulation time interval length, time interval [0,t]% -- f_s : [Hz] sampling frequency, set to 1000 if smaller.% Output(s)% -- Ts : [Sec][1xN double] time instances for the Rayleigh signal% -- z_dB : [dB] [1xN double] Rayleigh fading signal%

% Required parametersif f_s < 1000;f_s = 1000; % [Hz] Minumum required sampling rate

end;

N = ceil( t*f_s ); % Number of samples

% Ts contains the time instances at which z_dB is specifiedTs = linspace(0,t,N);

if mod( N, 2) == 1N = N+1; % Use even number of samples in calculation

end;f = linspace(-f_s,f_s,N); % [Hz] Frequency samples used in calculation

% Generate complex Gaussian samples with line spectra in frequency domain% Inphase :Gfi_p = randn(2,N/2); CGfi_p = Gfi_p(1,:)+i*Gfi_p(2,:); CGfi = [conj(fliplr(CGfi_p)) CGfi_p ];

% Quadrature :Gfq_p = randn(2,N/2); CGfq_p = Gfq_p(1,:)+i*Gfq_p(2,:); CGfq = [conj(fliplr(CGfq_p)) CGfq_p ];

% Generate fading spectrum, this is used to shape the Gaussian line spectraomega_p = 1; % this makes sure that the average received envelop can be 0dBS_r = omega_p/4/pi./(f_D*sqrt(1-(f/f_D).^2));

% Take care of samples outside the Doppler frequency range, let them be 0idx1 = find(f>=f_D); idx2 = find(f<=-f_D); S_r(idx1) = 0;S_r(idx2) = 0; S_r(idx1(1)) = S_r(idx1(1)-1);S_r(idx2(length(idx2))) = S_r(idx2(length(idx2))+1);

% Generate r_I(t) and r_Q(t) using inverse FFT:r_I = N*ifft(CGfi.*sqrt(S_r)); r_Q = -i*N*ifft(CGfq.*sqrt(S_r));

% Finally, generate the Rayleigh distributed signal envelopez = sqrt(abs(r_I).^2+abs(r_Q).^2); z_dB = 20*log10(z);

% Return correct number of pointsz_dB = z_dB(1:length(Ts));

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2−30

−20

−10

0 1 Hz

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2−30

−20

−10

0

10 10Hz

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2−60

−40

−20

0

20l00 Hz

Figure 3: Problem 11

12. Pr = 30dBmfD = 10Hz

P0 = 0dBm, tz =eρ2 − 1ρfD

√2π

= 0.0013s

P0 = 15dBm, tz = 0.0072s

P0 = 25dBm, tz = 0.0264s

13. In the reader, we found the level crossing rate below a level by taking an average of the number oftimes the level was crossed over a large period of time. It is easy to convince that the level crossingrate above a level will have the same expression as eq. 3.44 in reader because to go below a level again,we first need to go above it. There will be some discrepancy at the end points, but for a large enoughT it will not effect the result. So we have

LZ(above) = LZ(below) =√

2πfDρe−ρ2

And,

tZ(above) =p(z > Z)LZ(above)

p(z > Z) = 1− p(z ≤ Z) = 1− (1− e−ρ2) = e−ρ2

tZ(above) =1√

2πfDρ

The values of tZ(above) for fD = 10,50,80 Hz are 0.0224s, 0.0045s, 0.0028s respectively. Notice that asfD increases, tZ(above) reduces.

14. Note: The problem has been solved for Ts = 10µsPr = 10dBfD = 80Hz

R1 : −∞ ≤ γ ≤ −10dB, π1 = 9.95× 10−3

R2 : −10dB ≤ γ ≤ 0dB, π2 = 0.085R3 : 0dB ≤ γ ≤ 5dB, π3 = 0.176R4 : 5dB ≤ γ ≤ 10dB, π4 = 0.361R5 : 10dB ≤ γ ≤ 15dB, π5 = 0.325R6 : 15dB ≤ γ ≤ 20dB, π6 = 0.042R7 : 20dB ≤ γ ≤ 30dB, π7 = 4.54× 10−5

R8 : 30dB ≤ γ ≤ ∞, π8 = 3.72× 10−44

Nj → level crossing rate at level Aj

N1 = 0, ρ =√

010

N2 = 19.85, ρ =√

0.110

N3 = 57.38, ρ =√

110

N4 = 82.19, ρ =√

100.5

10

N5 = 73.77, ρ =√

1010

N6 = 15.09, ρ =√

101.5

10

N7 = 0.03, ρ =√

102

10

N8 = 0, ρ =√

103

10

MATLAB CODE:N = [0 19.85 57.38 82.19 73.77 15.09 .03 0];

Pi = [9.95e-3 .085 .176 .361 .325 .042 4.54e-5 3.72e-44];

T = 10e-3;for i = 1:8

if i == 1p(i,1) = 0;p(i,2) = (N(i+1)*T)/Pi(i);p(i,3) = 1-(p(i,1)+p(i,2));

elseif i == 8p(i,1) = (N(i)*T)/Pi(i);p(i,2) = 0;p(i,3) = 1-(p(i,1)+p(i,2));

elsep(i,1) = (N(i)*T)/Pi(i);p(i,2) = (N(i+1)*T)/Pi(i);p(i,3) = 1-(p(i,1)+p(i,2));

endend

% p =%% 0 0.0199 0.9801% 0.0023 0.0068 0.9909% 0.0033 0.0047 0.9921% 0.0023 0.0020 0.9957% 0.0023 0.0005 0.9973% 0.0036 0.0000 0.9964% 0.0066 0 0.9934% 0 0 1.0000

15. (a)

S(τ, ρ) =

α1δ(τ) ρ = 70Hzα2δ(τ − 0.022µsec) ρ = 49.5Hz0 else

The antenna setup is shown in Fig. 15From the figure, the distance travelled by the LOS ray is d and the distance travelled by the firstmultipath component is

2

√(d

2

)2

+ 64

Given this setup, we can plot the arrival of the LOS ray and the multipath ray that bounces offthe ground on a time axis as shown in Fig. 15So we have

2

√(d

2

)2

+ 82 − d = 0.022× 10−63× 108

⇒ 4(

d2

4+ 82

)= 6.62 + d2 + 2d(6.6)

⇒ d = 16.1m

fD = v cos(θ)/λ. v = fDλ/ cos(θ). For the LOS ray, θ = 0 and for the multipath component,θ = 45o. We can use either of these rays and the corresponding fD value to get v = 23.33m/s.

(b)

dc =4hthr

λ

dc = 768 m. Since d ¿ dc, power fall-off is proportional to d−2.

(c) Tm = 0.022µs, B−1 = 0.33µs. Since Tm ¿ B−1, we have flat fading.

16. (a) Outdoor, since delay spread ≈ 10 µsec.Consider that 10 µsec ⇒ d = ct = 3km difference between length of first and last path

(b) Scattering functionS(τ, ρ) = F∆t[Ac(τ, ∆t)]

= 1W rect

(1W ρ

)for 0 ≤ τ ≤ 10µsec

The Scattering function is plotted in Fig. 16

8m 8m

d meters

Figure 4: Antenna Setting

0.022 us

0 t

t0 = (d/3e8)

t1 = 2 sqrt[(d/2)^2+8^2]/3e8

t1t0

Figure 5: Time Axis for Ray Arrival

(c) Avg Delay Spread =

∞R0

τAc(τ)dτ

∞R0

Ac(τ)dτ= 5µsec

RMS Delay Spread =

√√√√√∞R0

(τ−µTm)2Ac(τ)dτ

∞R0

Ac(τ)dτ= 2.89µsec

Doppler Spread = W2 = 50 Hz

(d) βu > Coherence BW ⇒ Freq. Selective Fading ≈ 1Tm

= 105 ⇒ βu > = 105 kHzCan also use µTm or σTm instead of Tm

(e) Rayleigh fading, since receiver power is evenly distributed relative to delay; no dominant LOSpath

(f) tR = eρ2−1ρFD

√2π

with ρ = 1, fD = W2 → tr = .0137 sec

(g) Notice that the fade duration never becomes more than twice the average. So, if we choose ourdata rate such that a single symbol spans the average fade duration, in the worst case two symbolswill span the fade duration. So our code can correct for the lost symbols and we will have error-freetransmission. So 1

tR= 72.94 symbols/sec

17. (a) Tm ≈ .1msec = 100µsecBd ≈ .1HzAnswers based on µTm or σTm are fine too. Notice, that based on the choice of either Tm, µTm orσTm , the remaining answers will be different too.

(b) Bc ≈ 1Tm

= 10kHz∆f > 10kHz for u1 ⊥ u2

(c) (∆t)c = 10s

(d) 3kHz < Bc ⇒ Flat30kHz > Bc ⇒ Freq. Selective

rho

tau

W/2-W/2

10 us

S(tau,rho)

Figure 6: Scattering Function

Chapter 4

1. C = B log2

(1 + S

N0B

)

C =log2

1+ S

N0B

1B

As B →∞ by L’Hospital’s rule

C =S

N0

1ln 2

2. B = 50 MHzP = 10 mWN0 = 2 ×10−9 W/HzN = N0BC = 6.87 Mbps.

Pnew = 20 mW, C = 13.15 Mbps (for x ¿ 1, log(1 + x) ≈ x)B = 100 MHz, Notice that both the bandwidth and noise power will increase. So C = 7 Mbps.

3. Pnoise = 0.1mWB = 20MHz

(a) Cuser1→base station = 0.933B = 18.66Mbps

(b) Cuser2→base station = 3.46B = 69.2Mbps

4. (a) Ergodic Capacity (with Rcvr CSI only)= B[∑6

i=1 log2(1 + γi)p(γi)] = 2.8831×B = 57.66 Mbps.

(b) pout = Pr(γ < γmin)Co = (1-pout)Blog2(1 + γmin)Forγmin > 20dB, pout = 1, Co = 015dB < γmin < 20dB, pout = .9, Co = 0.1Blog2(1 + γmin), max Co at γmin ≈ 20dB.10dB < γmin < 15dB, pout = .75, Co = 0.25Blog2(1 + γmin), max Co at γmin ≈ 15dB.5dB < γmin < 10dB, pout = .5, Co = 0.5Blog2(1 + γmin), max Co at γmin ≈ 10dB.0dB < γmin < 5dB, pout = .35, Co = 0.65Blog2(1 + γmin), max Co at γmin ≈ 5dB.−5dB < γmin < 0dB, pout = .1, Co = 0.9Blog2(1 + γmin), max Co at γmin ≈ 0dB.γmin < −5dB, pout = 0, Co = Blog2(1 + γmin), max Co at γmin ≈ -5dB.

Plot is shown in Fig. 1. Maximum at γmin = 10dB, pout=0.5 and Co = 34.59 Mbps.

5. (a) We suppose that all channel states are used

1γ0

= 1 +4∑

i=1

1γi

pi ⇒ γ0 = 0.8109

1γ0− 1

γ4> 0 ∴ true

S(γi)S

=1γ0− 1

γi

0 0.2 0.4 0.6 0.8 10

0.5

1

1.5

2

2.5

3

3.5x 10

7

Pout

Cap

acity

(bp

s)

Figure 1: Capacity vs Pout

S(γ)S

=

1.2322 γ = γ1

1.2232 γ = γ2

1.1332 γ = γ3

0.2332 γ = γ4

C

B=

4∑

i=1

log2

(γi

γ0

)p(γi) = 5.2853bps/Hz

(b) σ = 1E[1/γ] = 4.2882

S(γi)

S= σ

γi

S(γ)S

=

0.0043 γ = γ1

0.0029 γ = γ2

0.4288 γ = γ3

4.2882 γ = γ4

C

B= log2(1 + σ) = 2.4028bps/Hz

(c) To have pout = 0.1 or 0.01 we will have to use all the sub-channels as leaving any of these willresult in a pout of at least 0.2 ∴ truncated channel power control policy and associated spectralefficiency are the same as the zero-outage case in part b .To have pout that maximizes C with truncated channel inversion, we get

maxC

B= 4.1462bps/Hz pout = 0.5

6. (a)

SNRrecvd =Pγ(d)Pnoise

=

10dB w.p. 0.45dB w.p. 0.30dB w.p. 0.2

−10dB w.p. 0.1

Assume all channel states are used

1γ0

= 1 +4∑

i=1

1γi

pi ⇒ γ0 = 0.4283 > 0.1 ∴ not possible

Now assume only the best 3 channel states are used

0.9γ0

= 1 +3∑

i=1

1γi

pi ⇒ γ0 = 0.6742 < 1 ∴ ok!

S(γ)S

=

1.3832 γ = γ1 = 101.1670 γ = γ2 = 3.16230.4832 γ = γ3 = 10 γ = γ4 = 0.1

C/B = 2.3389bps/Hz

(b) σ = 0.7491C/B = log2(1 + σ) = 0.8066bps/Hz

(c)(

CB

)max

= 2.1510bps/Hz obtained by using the best 2 channel states.With pout = 0.1 + 0.2 = 0.3

7. (a) Maximize capacity given by

C = maxS(γ):

RS(γ)p(γ)dγ=S

∫

γB log

(1 +

S(γ)γS

)p(γ)dγ.

Construct the Lagrangian function

L =∫

γB log

(1 +

S(γ)γS

)p(γ)dγ − λ

∫S(γ)

Sp(γ)dγ

Taking derivative with respect to S(γ), (refer to discussion section notes) and setting it to zero,we obtain,

S(γ)S

= 1

γ0− 1

γ γ ≥ γ0

0 γ < γ0

Now, the threshold value must satisfy∫ ∞

γ0

(1γ0− 1

γ

)p(γ)dγ = 1

Evaluating this with p(γ) = 110e−γ/10, we have

1 =1

10γ0

∫ ∞

γ0

e−γ/10dγ − 110

∫ ∞

γ0

e−γ/10

γdγ (1)

=1γ0

e−γ0/10 − 110

∫ ∞

γ010

e−γ

γdγ (2)

=1γ0

e−γ0/10 − 110

EXPINT(γ0/10) (3)

where EXPINT is as defined in matlab. This gives γ0 = 0.7676. The power adaptation becomes

S(γ)S

= 1

0.7676 − 1γ γ ≥ 0.7676

0 γ < 0.7676

(b) Capacity can be computed as

C/B =110

∫ ∞

0.7676log (γ/0.7676) e−γ/10dγ = 2.0649 nats/sec/Hz.

Note that I computed all capacites in nats/sec/Hz. This is because I took the natural log. Inorder to get the capacity values in bits/sec/Hz, the capacity numbers simply need to be dividedby natural log of 2.

(c) AWGN capacity C/B = log(1 + 10) = 2.3979 nats/sec/Hz.

(d) Capacity when only receiver knows γ

C/B =110

∫ ∞

0log (1 + γ) e−γ/10dγ = 2.0150 nats/sec/Hz.

(e) Capacity using channel inversion is ZERO because the channel can not be inverted with finiteaverage power. Threshold for outage probability 0.05 is computed as

110

∫ ∞

γ0e−γ/10dγ = 0.95

which gives γ0 = 0.5129. This gives us the capacity with truncated channel inversion as

C/B = log

[1 +

1110

∫∞γ0

1γ e−γ/10dγ

]∗ 0.95 (4)

= log

[1 +

1110EXPINT(γ0/10)

]∗ 0.95 (5)

= 1.5463 nats/s/Hz. (6)

(f) Channel Mean=-5 dB = 0.3162. So for perfect channel knowledge at transmitter and receiver wecompute γ0 = 0.22765 which gives capacity C/B = 0.36 nats/sec/Hz.With AWGN, C/B = log(1 + 0.3162) = 0.2748 nats/sec/Hz.With channel known only to the receiver C/B = 0.2510 nats/sec/Hz.

Capacity with AWGN is always greater than or equal to the capacity when only the recieverknows the channel. This can be shown using Jensen’s inequality. However the capacity whenthe transmitter knows the channel as well and can adapt its power, can be higher than AWGNcapacity specially at low SNR. At low SNR, the knowledge of fading helps to use the low SNRmore efficiently.

8. (a) If neither transmitter nor receiver knows when the interferer is on, they must transmit assumingworst case, i.e. as if the interferer was on all the time,

C = B log(

1 +S

N0B + I

)= 10.7Kbps.

(b) Suppose we transmit at power S1 when jammer is off and S2 when jammer is off,

C = B max[log

(1 +

S1

NoB

)0.75 + log

(1 +

S2

NoB + I

)0.25

]

subject to0.75S1 + 0.25S2 = S.

This gives S1 = 12.25mW , S2 = 3.25mW and C = 53.21Kbps.

(c) The jammer should transmit −x(t) to completely cancel off the signal.

S = 10mWN0 = .001 µW/HzB = 10 MHz

Now we compute the SNR’s as:

γj =|Hj |2SN0B

This gives: γ1 = |1|210−3

0.001×10−610×106 = 1, γ2 = .25, γ3 = 4, γ4 = 0.0625

To compute γ0 we use the power constraint as:∑

j

(1γ0− 1

γj

)

+

= 1

First assume that γ0 < 0.0625, then we have

4γ0

= 1 +(

11

+1

.25+

14

+1

.0625

)

⇒ γ0 = .1798 > 0.0625

So, our assumption was wrong. Now we assume that 0.0625 < γ0 < .25, then

3γ0

= 1 +(

11

+1

.25+

14

)

⇒ γ0 = .48 > 0.25

So, our assumption was wrong again. Next we assume that 0.25 < γ0 < 1, then

2γ0

= 1 +(

11

+14

)

⇒ γ0 = .8889 < 1

This time our assumption was right. So we get that only two sub-bands each of bandwidth 10 MHzare used for transmission and the remaining two with lesser SNR’s are left unused.

Now, we can find capacity as:

C =∑

j:γj≥γ0

B log2

(γj

γ0

)

This gives us, C = 23.4 Mbps.

9. Suppose transmit power is Pt, interference power is Pint and noise power is Pnoise.In the first strategy C/B = log2

(1 + Pt

Pint+Pnoise

)

In the second strategy C/B = log2

(1 + Pt−Pint

Pnoise

)

Assuming that the transmitter transmits -x[k] added to its message always, power remaining for actualmessages is Pt − Pint

The first or second strategy may be better depending on

Pt

Pint + Pnoise≷ Pt − Pint

Pnoise⇒ Pt − Pint − Pnoise ≷ 0

Pt is generally greater than Pint + Pnoise , and so strategy 2 is usually better.

1

0.5

2

0.25

f (in MHz)

fcfc+10

fc+20fc-10

fc-20

H(f)

Figure 2: Problem 11

10. We show this for the case of a discrete fading distribution

C = Σ log(

1 +(1 + j)2Pj

N0B

)

L =∑

i

log(

1 +(1 + j)2Pj

N0B

)− dj

∑

j

Pj − P

∂L∂Pj

= 0

⇒ (1 + j)2Pj

N0B=

1λ

(1 + j)2

N0B− 1

letγj =(1 + j)2P

N0B

⇒ Pj

P=

1λP

− 1γj

denote1γ0

=1

λP

∴ Pj

P=

1γ0− 1

γj

subject to the constraintΣPj

P= 1

11. S = 10mWN0 = .001 µW/HzB = 10 MHz

Now we compute the SNR’s as:

γj =|Hj |2SN0B

This gives: γ1 = |1|210−3

0.001×10−610×106 = 1, γ2 = .25, γ3 = 4, γ4 = 0.0625

To compute γ0 we use the power constraint as:∑

j

(1γ0− 1

γj

)

+

= 1

First assume that γ0 < 0.0625, then we have

4γ0

= 1 +(

11

+1

.25+

14

+1

.0625

)

⇒ γ0 = .1798 > 0.0625

So, our assumption was wrong. Now we assume that 0.0625 < γ0 < .25, then

3γ0

= 1 +(

11

+1

.25+

14

)

⇒ γ0 = .48 > 0.25

So, our assumption was wrong again. Next we assume that 0.25 < γ0 < 1, then

2γ0

= 1 +(

11

+14

)

⇒ γ0 = .8889 < 1

This time our assumption was right. So we get that only two sub-bands each of bandwidth 10 MHzare used for transmission and the remaining two with lesser SNR’s are left unused.

Now, we can find capacity as:

C =∑

j:γj≥γ0

B log2

(γj

γ0

)

This gives us, C = 23.4 Mbps.

12. For the case of a discrete number of frequency bands each with a flat frequency response, the problemcan be stated as

maxs.t.P

i P (fi)≤P

∑

i

log2

(1 +

|H(fi)|2P (fi)N0

)

denote γ(fi) = |H(fi)|2P (fi)N0

L =∑

i

log2

(1 + γ(fi)

P (fi)P

)+ λ(

∑P (fi))

denote xi = P (fi)P , the problem is similar to problem 10

⇒ x?i =

1γ0− 1

γ(fi)

where γ0 is found from the constraints

∑

i

(1γ0− 1

γ(fi)

)= 1 and

1γ0− 1

γ(fi)≥ 0∀i

13. (a) C=13.98Mbps

MATLAB

Gammabar = [1 .5 .125];ss = .001;P = 30e-3;N0 = .001e-6;

Bc = 4e6;Pnoise = N0*Bc;hsquare = [ss:ss:10*max(Gammabar)];gamma = hsquare*(P/Pnoise);

for i = 1:length(Gammabar)pgamma(i,:) = (1/Gammabar(i))*exp(-hsquare/Gammabar(i));

end

gamma0v = [1:.01:2];for j = 1:length(gamma0v)

gamma0 = gamma0v(j);sumP(j) = 0;for i = 1:length(Gammabar)

a = gamma.*(gamma>gamma0);[b,c] = max(a>0);gammac = a(find(a));pgammac = pgamma(i,c:length(gamma));Pj_by_P = (1/gamma0)-(1./gammac);sumP(j) = sumP(j) + sum(Pj_by_P.*pgammac)*ss;

endend[b,c] = min(abs((sumP-1)));gamma0ch = gamma0v(c);

C = 0;for i = 1:length(Gammabar)

a = gamma.*(gamma>gamma0ch);[b,c] = max(a>0);gammac = a(find(a));pgammac = pgamma(i,c:length(gamma));C = C + Bc*ss*sum(log2(gammac/gamma0ch).*pgammac);

end

(b) C=13.27Mbps

MATLAB

Gammabarv = [1 .5 .125];ss = .001;Pt = 30e-3;N0 = .001e-6;

Bc = 4e6;Pnoise = N0*Bc;

P = Pt/3;for k = 1:length(Gammabarv)

Gammabar = Gammabarv(k);hsquare = [ss:ss:10*Gammabar];gamma = hsquare*(P/Pnoise);pgamma = (1/Gammabar)*exp(-hsquare/Gammabar);gamma0v = [.01:.01:1];for j = 1:length(gamma0v)

gamma0 = gamma0v(j);a = gamma.*(gamma>gamma0);[b,c] = max(a>0);gammac = a(find(a));pgammac = pgamma(c:length(gamma));Pj_by_P = (1/gamma0)-(1./gammac);sumP(j) = sum(Pj_by_P.*pgammac)*ss;

end[b,c] = min(abs((sumP-1)));gamma0ch = gamma0v(c);a = gamma.*(gamma>gamma0ch);[b,c] = max(a>0);gammac = a(find(a));pgammac = pgamma(c:length(gamma));C(k) = Bc*ss*sum(log2(gammac/gamma0ch).*pgammac);

end Ctot = sum(C);

Chapter 6

1. (a) For sinc pulse, B = 12Ts

⇒ Ts = 12B = 5× 10−5s

(b) SNR = PbN0B = 10

Since 4-QAM is multilevel signallingSNR = Pb

N0B = EsN0BTs

= 2EsN0B

(∵ BTs = 1

2

)

∴ SNR per symbol = EsN0

= 5SNR per bit = Eb

N0= 2.5 (a symbol has 2 bits in 4QAM)

(c) SNR per symbol remains the same as before = EsN0

= 5SNR per bit is halved as now there are 4 bits in a symbol Eb

N0= 1.25

2. p0 = 0.3, p1 = 0.7

(a)

Pe = Pr(0 detected, 1 sent — 1 sent)p(1 sent) + Pr(1 detected, 0 sent — 0 sent)p(0 sent)

= 0.7Q

(dmin√2N0

)+ 0.3Q

(dmin√2N0

)= Q

(dmin√2N0

)

dmin = 2A

= Q

√2A2

N0

(b)p(m = 0|m = 1)p(m = 1) = p(m = 1|m = 0)p(m = 0)

0.7Q

A + a√

N02

= 0.3Q

A− a√

N02

, a > 0

Solving gives us ’a’ for a given A and N0

(c)p(m = 0|m = 1)p(m = 1) = p(m = 1|m = 0)p(m = 0)

0.7Q

A√

N02

= 0.3Q

B√

N02

, a > 0

Clearly A > B, for a given A we can find B

(d) Take EbN0

= A2

N0= 10

In part a) Pe = 3.87× 10−6

In part b) a=0.0203 Pe = 3.53× 10−6

In part c) B=0.9587 Pe = 5.42× 10−6

Clearly part (b) is the best way to decode.

MATLAB CODE:A = 1;N0 = .1;a = [0:.00001:1];t1 = .7*Q(A/sqrt(N0/2));

t2=.3*Q(a/sqrt(N0/2));diff = abs(t1-t2);[c,d] = min(diff);a(d)c

3. s(t) = ±g(t) cos 2πfctr = r cos∆φwhere r is the signal after the sampler if there was no phase offset. Once again, the threshold thatminimizes Pe is 0 as (cos ∆φ) acts as a scaling factor for both +1 and -1 levels. Pe however increasesas numerator is reduced due to multiplication by cos ∆φ

Pe = Q

(dmin cos ∆φ√

2N0

)

4.

A2c

∫ Tb

0cos2 2πfctdt = A2

c

∫ Tb

0

1 + cos 4πfct

2

= A2c

Tb

2+

sin(4πfcTb)8πfc︸ ︷︷ ︸

→0 as fcÀ1

=A2

cTb

2= 1

x(t) = 1 + n(t)Let prob 1 sent =p1 and prob 0 sent =p0

Pe =16[1.p1 + 0.p0] +

26[0.p1 + 0.p0] +

26[0.p1 + 0.p0] +

16[0.p1 + 1.p0]

=16[p1 + p0] =

16

(∵ p1 + p0 = 1 always )

5. We will use the approximation Pe ∼ (average number of nearest neighbors).Q(

dmin√2N0

)

where number of nearest neighbors = total number of points taht share decision boundary

(a) 12 inner points have 5 neighbors4 outer points have 3 neighborsavg number of neighbors = 4.5Pe = 4.5Q

(2a√2N0

)

(b) 16QAM, Pe = 4(1− 1

4

)Q

(2a√2N0

)= 3Q

(2a√2N0

)

(c) Pe ∼ 2×3+3×25 Q

(2a√2N0

)= 2.4Q

(2a√2N0

)

(d) Pe ∼ 1×4+4×3+4×29 Q

(3a√2N0

)= 2.67Q

(3a√2N0

)

6.

Ps,exact = 1−(

1− 2(√

M − 1)√M

Q

(√3γs

M − 1

))2

Figure 1: Problem 5

Ps,approx =4(√

M − 1)√M

Q

(√3γs

M − 1

)

approximation is better for high SNRs as then the multiplication factor is not important and Pe isdictated by the coefficient of the Q function which are same.

MATLAB CODE:gamma_db = [0:.01:25];gamma = 10.^(gamma_db/10);M = 16;

Ps_exact=1-exp(2*log((1-((2*(sqrt(M)-1))/(sqrt(M)))*Q(sqrt((3*gamma)/(M-1))))));Ps_approx = ((4*(sqrt(M)-1))/sqrt(M))*Q(sqrt((3*gamma)/(M-1)));semilogy(gamma_db, Ps_exact);hold onsemilogy(gamma_db,Ps_approx,’b:’);

7. See figure. The approximation error decreases with SNR because the approximate formula is basedon nearest neighbor approximation which becomes more realistic at higher SNR. The nearest neighborbound over-estimates the error rate because it over-counts the probability that the transmitted signalis mistaken for something other than its nearest neighbors. At high SNR, this is very unlikely and thisover-counting becomes negligible.

8. (a)

Ix(a) =∫ ∞

0

e−at2

x2 + t2dt

since the integral converges we can interchange integral and derivative for a¿0

∂Ix(a)∂a

=∫ ∞

0

−te−at2

x2 + t2dt

x2Ix(a)− ∂Ix(a)∂a

=∫ ∞

0

(x2 + t2)e−at2

x2 + t2dt =

∫ ∞

0e−at2dt =

12

√π

a

0 5 10 15 20 25 3010

−300

10−200

10−100

100

Ps

Problem 2 − Symbol Error Probability for QPSK

ApproximationExact Formula

0 1 2 3 4 5 6 7 8 9 1010

−3

10−2

10−1

100

SNR(dB)

Ps

Problem 2 − Symbol Error Probability for QPSK

ApproximationExact Formula

Figure 2: Problem 7

(b) Let Ix(a) = y, we get

y′ − x2y = −12

√π

a

comparing withy′ + P (a)y = Q(a)

P (a) = −x2 , Q(a) = −12

√π

a

I.F. = eR

P (u)u = e−x2a

∴ e−x2ay =∫−1

2

√π

ae−x2udu

solving we get

y =π

2xeax2

erfc(x√

a)

(c)

erfc(x√

a) = Ix(a)2x

πe−ax2

=2x

πe−ax2

∫ ∞

0

e−at2

x2 + t2dt

a = 1

erfc(x) =2x

πe−ax2

∫ ∞

0

e−at2

x2 + t2dt

=2π

∫ π/2

0e−x2/sin2θdθ

Q(x) =12erfc(x/

√2) =

1π

∫ π/2

0e−x2/2sin2θdθ

9. P = 100WN0 = 4W, SNR = 25Pe = Q(

√2γ) = Q(

√50) = 7.687× 10−13

data requires Pe ∼ 10−6

voice requires Pe ∼ 10−3

so it can be used for both.with fading Pe = 1

4γb= 0.01

So the system can’t be used for data at all. It can be used for very low quality voice.

10. Ts = 15µsecat 1mph Tc = 1

Bd= 1

v/λ = 0.74s À Ts

∴ outage probability is a good measure.at 10 mph Tc = 0.074s À Ts ∴ outage probability is a good measure.at 100 mph Tc = 0.0074s = 7400µs > 15µs outage or outage combined with average prob of error canbe a good measure.

11.

Mγ(s) =∫ ∞

0esγp(γ)dγ

=∫ ∞

0esγ 1

γe−γ/γdγ

=1

1− γs

12. (a) When there is path loss alone, d =√

1002 + 5002 = 100√

6× 103

Pe = 12e−γb ⇒ γb = 13.1224

Pγ

N0B = 13.1224 ⇒ Pγ = 1.3122× 10−14

Pγ

Pt=

[√Gλ

4πd

]2⇒ 4.8488W

(b)x = 1.3122× 10−14 = −138.82dB

Pγ,dB ∼ N(µPγ , 8), σdB = 8

P (Pγ,dB ≥ x) = 0.9

P

(Pγ,dB − µPγ

8≥ x− µPγ

8

)= 0.9

⇒ Q

(x− µPγ

8

)= 0.9

⇒ x− µPγ

8= −1.2816

⇒ µPγ = −128.5672dB = 1.39× 10−13

13. (a) Law of Cosines:c2 = a2 + b2 − 2ab cosC with a,b =

√Es, c = dmin, C = Θ = 22.5

c = dmin =√

2Es(1− cos 22.5) = .39√

Es

Can also use formula from reader

(b) Ps = αmQ(√

βmγs

)= 2Q

(√dmin

2

2No

)= 2Q(

√.076γs)

αm = 2, βm = .076

(c) Pe =∞∫0

Ps(γs)f(γs)dγs

=∞∫0

αmQ(√

βmγs)f(γs)dγs

Using alternative Q form

= αmπ

π2∫0

(1 + gγs

(sin φ)2

)−1dφ with g = βm

2

= αm2

[1−

√gγs

1+gγs

]= 1−

√.038γs

1+.038γs= 1

.076γs, where we have used an integral table to evaluate

the integral

(d) Pd = Ps4

(e) BPSK: P b = 14γb

= 10−3, ⇒ γb = 250, 16PSK: From above get γs = 3289.5Penalty = 3289.5

250 = 11.2dBAlso will accept γb(16PSK) = 822 ⇒= 5.2dB

14.

P b =∫ ∞

0Pb(γ)p(γ)dγ

Pb(γ) =12e−γ

P b =12

∫ ∞

0e−γbpγ(γ)dγ =

12M

But from 6.65

Mγ(s) =(1− sγ

m

)−m

∴ P b =12

(1 +

γ

m

)−m

For M = 4, γ = 10P b = 3.33× 10−3

15. %Script used to plot the average probability of bit error for BPSK modulation in%Nakagami fading m = 1, 2, 4.%Initializationsavg_SNR = [0:0.1:20]; gamma_b_bar = 10.^(avg_SNR/10); m = [1 2 4];line = [’-k’, ’-r’, ’-b’]

for i = 1:size(m,2)for j = 1:size(gamma_b_bar, 2)

Pb_bar(i,j) = (1/pi)*quad8(’nakag_MGF’,0,pi/2,[],[],gamma_b_bar(j),m(i),1);endfigure(1);semilogy(avg_SNR, Pb_bar(i,:), line(i));hold on;

end

xlabel(’Average SNR ( gamma_b ) in dB’); ylabel(’Average bit errorprobability ( P_b ) ’); title(’Plots of P_b for BPSK modulation inNagakami fading for m = 1, 2, 4’); legend(’m = 1’, ’m = 2’, ’m =4’);

function out = nakag_MGF(phi, gamma_b_bar, m, g);%This function calculates the m-Nakagami MGF function for the specified values of phi.%phi can be a vector. Gamma_b_bar is the average SNR per bit, m is the Nakagami parameter%and g is given by Pb(gamma_b) = aQ(sqrt(2*g*gamma_b)).

out = (1 + gamma_b_bar./(m*(sin(phi).^2)) ).^(-m);

SNR = 10dB

M BER1 2.33×10−2

2 5.53×10−3

4 1.03×10−3

16. For DPSK in Rayleigh fading, Pb = 12γb

⇒ γb = 500NoB = 3× 10−12mW ⇒ Ptarget = γbN0B = 1.5× 10−9mW = -88.24 dBm

Now, consider shadowing:Pout = P [Pr < Ptarget] = P [Ψ < Ptarget − Pr] = Φ

(Ptarget−Pr

σ

)

⇒ Φ−1(.01) = 2.327 = Ptarget−Pr

σ

Pr = −74.28 dBm = 3.73× 10−8 mW = Pt

(λ

4πd

)2

⇒ d = 1372.4 m

17. (a)

γ1 =

0 w.p. 1/330 w.p. 2/3

γ2 =

5 w.p. 1/210 w.p. 1/2

In MRC, γΣ = γ1+γ2. So,

γΣ =

5 w.p. 1/610 w.p. 1/635 w.p. 1/340 w.p. 1/3

(b) Optimal Strategy is water-filling with power adaptation:

S(γ)S

= 1

γ0− 1

γ , γ ≥ γ0

0 γ < γ0

Notice that we will denote γΣ by γ only hereon to lighten notation. We first assume γ0 < 5,

4∑

i=1

(1γ0− 1

γi

)pi = 1

⇒ 1γ0

= 1 +4∑

i=1

pi

γi

⇒ γ0 = 0.9365 < 5

So we found the correct value of γ0.

C = B

4∑

i=1

log2

(γi

γ0

)pi

C = 451.91 Kbps

(c) Without, receiver knowledge, the capacity of the channel is given by:

C = B

4∑

i=1

log2(1 + γi)pi

C = 451.66 KbpsNotice that we have denote γΣ by γ to lighten notation.

18. (a)

s(k) = s(k − 1)z(k − 1) = gk−1s(k − 1) + n(k − 1)

z(k) = gks(k) + n(k)

From equation 5.63 , the input to the phase comparator is

z(k)z?(k − 1) = gkg(k − 1)?s(k)s?(k − 1) + gks(k − 1)n?k−1 +

g(k − 1)?s?(k − 1)nk + nkn?k−1

but s(k) = s(k − 1)s(k)s?(k − 1) = |sk|2 = 1 (normalized)

(b)

nk = s?k−1nk

nk−1 = s?k−1nk−1

z = gkg?k−1 + gkn

?k−1 + g?

k−1nk

φx(s) =p1p2

(s− p1)(s− p2)=

A

s− p1+

B

s− p2

A = (s− p1)φx(s)|s=p1 =p1p2

p1 − p2

B = (s− p2)φx(s)|s=p2 =p1p2

p2 − p1

(c) Relevant part of the pdfφx(s) =

p1p2

(p2 − p1)(s− p2)

∴ px(x) =p1p2

(p2 − p1)L−1

(1

(s− p2)

)=

p1p2

(p2 − p1)ep2x , x < 0

(d)

Pb = prob(x < 0) =p1p2

(p2 − p1)

∫ 0

−∞ep2xdx = − p1

p2 − p1

(e)

p2 − p1 =1

2N0[γb(1− ρc) + 1]+

12N0[γb(1 + ρc) + 1]

=γb + 1

N0[γb(1− ρc) + 1][γb(1 + ρc) + 1]

∴ P b =γb(1− ρc) + 1

2(γb + 1)

(f) ρc = 1

∴ P b =1

2(γb + 1)

19. γb 0 to 60dBρc = J0(2πBDT ) with BDT = 0.01, 0.001, 0.0001where J0 is 0 order Bessel function of 1st kind.

P b =12

[1 + γb(1− ρc)

1 + γb

]

when BDT = 0.01, floor can be seen about γb = 40dBwhen BDT = 0.001, floor can be seen about γb = 60dBwhen BDT = 0.0001, floor can be seen between γb = 0 to 60dB

20. Data rate = 40 KbpsSince DQPSK has 2 bits per symbol. ∴ Ts = 2

40×103 = 5× 10−5sec

DQPSKGaussian Doppler power spectrum, ρc = e−(πBDT )2

BD = 80HzRician fading K = 2ρc = 0.9998

P floor =12

[1−

√(ρc/

√2)2

1− (ρc/√

2)2

]exp

[−(2−√2)K/2

1− ρc/√

2

]= 2.138× 10−5

21. ISI:

Formula based approach:

Pfloor =(

σTm

Ts

)2

Since its Rayleigh fading, we can assume that σTm ≈ µTm = 100nsPfloor ≤ 10−4

which gives us

(σTm

Ts

)2

≤ 10−4

Ts ≥ σTm√Pfloor

= 10µsec

So, Ts ≥ 10µs. Tb ≥ 5µs. Rb ≤ 200 Kbps.

Thumb - Rule approach:

µt = 100 nsec will determine ISI. As long as Ts À µT , ISI will be negligible. Let Ts ≥ 10 µT . ThenR ≤ 2bits

symbol1Ts

symbolssec = 2Mbps

Doppler:

BD = 80 Hz

Pfloor = 10−4 ≥ 12

1−

√√√√(ρc/√

2)2

1− (ρc/√

2)2

⇒ ρc ≥ 0.9999

But ρc for uniform scattering is J0(2πBDTs), so

ρc = J0(2πBDTs) = 1− (πfDTs)2 ≥ 0.9999

⇒ Ts ≤ 39.79µs

Tb ≤ 19.89µs. Rb ≥ 50.26 Kbps.

Combining the two, we have 50.26 Kbps ≤ Rb ≤ 200 Kbps (or 2 Mbps).

22. From figure 6.5with Pb = 10−3, d = θTm/Ts, θTm = 3µsBPSK

d = 5× 10−2

Ts = 60µsec

R = 1/Ts = 16.7Kbps

QPSKd = 4× 10−2

Ts = 75µsec

R = 2/Ts = 26.7Kbps

MSKd = 3× 10−2

Ts = 100µsec

R = 2/Ts = 20Kbps

Chapter 7

1. Ps = 10−3

QPSK, Ps = 2Q(√

γs) ≤ 10−3, γs ≥ γ0 = 10.8276.

Pout(γ0) =M∏

i=1

(1− e

− γ0γi

)

γ1 = 10, γ2 = 31.6228, γ3 = 100.

M = 1Pout =

(1− e

− γ0γ1

)= 0.6613

M = 2Pout =

(1− e

− γ0γ1

) (1− e

− γ0γ2

)= 0.1917

M = 3Pout =

(1− e

− γ0γ1

) (1− e

− γ0γ2

) (1− e

− γ0γ3

)= 0.0197

2. pγΣ(γ) = Mγ

[1− e−γ/γ

]M−1e−γ/γ

γ = 10dB = 10as we increase M, the mass in the pdf keeps on shifting to higher values of γ and so we have highervalues of γ and hence lower probability of error.

MATLAB CODE

gamma = [0:.1:60];gamma_bar = 10;M = [1 2 4 8 10];fori=1:length(M)

pgamma(i,:) = (M(i)/gamma_bar)*(1-exp(-gamma/gamma_bar)).^...(M(i)-1).*(exp(-gamma/gamma_bar));

end

3.

P b =∫ ∞

0

12e−γpγΣ(γ)dγ

=∫ ∞

0

12e−γ M

γ

[1− e−γ/γ

]M−1e−γ/γdγ

=M

2γ

∫ ∞

0e−(1+1/γ)γ

[1− e−γ/γ

]M−1dγ

=M

2γ

M−1∑

n=0

(M − 1n

)(−1)ne−(1+1/γ)γdγ

=M

2

M−1∑

n=0

(M − 1n

)(−1)n 1

1 + n + γ= desired expression

0 10 20 30 40 50 600

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

γ

p γ Σ(γ)

M = 1M = 2M = 4M = 8M = 10

Figure 1: Problem 2

4.

pγΣ(γ) =

Prγ2 < γτ , γ1 < γ γ < γτ

Prγτ ≤ γ1 ≤ γ+ Prγ2 < γτ , γ1 < γ γ > γτ

If the distribution is iid this reduces to

pγΣ(γ) =

Pγ1(γ)Pγ2(γτ ) γ < γτ

Prγτ ≤ γ1 ≤ γ+ Pγ1(γ)Pγ2(γτ ) γ > γτ

5.

P b =∫ ∞

0

12e−γpγΣ(γ)dγ

pγΣ(γ) =

(1− e−γT /γ

)1γ e−γr/γ γ < γT(

2− e−γT /γ)

1γ e−γr/γ γ > γT

P b =12γ

(1− e−γT /γ

)∫ γT

0e−γ/γe−γdγ +

12γ

(2− e−γr/γ

)∫ ∞

γT

e−γ/γe−γdγ

=1

2(γ + 1)

(1− e−γT /γ + e−γT e−γT /γ

)

6.P b P b(10dB) P b(20dB)

no diversity 12(γ+1) 0.0455 0.0050

SC(M=2) M2

∑M−1m=0 (−1)m

0@ M − 1

m

1A

1+m+γ 0.0076 9.7× 10−5

SSC 12(γ+1)

(1− e−γT /γ + e−γT e−γT /γ

)0.0129 2.7× 10−4

As SNR increases SSC approaches SC

7. See

MATLAB CODE:gammab_dB = [0:.1:20];gammab = 10.^(gammab_dB/10);M= 2;

0 2 4 6 8 10 12 14 16 18 2010

−7

10−6

10−5

10−4

10−3

10−2

10−1

100

γavg

Pb,

avg (

DP

SK

)

M = 2M = 3M = 4

Figure 2: Problem 7

for j = 1:length(gammab)Pbs(j) = 0for m = 0:M-1

f = factorial(M-1)/(factorial(m)*factorial(M-1-m));Pbs(j) = Pbs(j) + (M/2)*((-1)^m)*f*(1/(1+m+gammab(j)));

endendsemilogy(gammab_dB,Pbs,’b--’)hold on

M = 3;for j = 1:length(gammab)

Pbs(j) = 0for m = 0:M-1

f = factorial(M-1)/(factorial(m)*factorial(M-1-m));Pbs(j) = Pbs(j) + (M/2)*((-1)^m)*f*(1/(1+m+gammab(j)));

endendsemilogy(gammab_dB,Pbs,’b-.’);hold on

M = 4;for j = 1:length(gammab)

Pbs(j) = 0for m = 0:M-1

f = factorial(M-1)/(factorial(m)*factorial(M-1-m));Pbs(j) = Pbs(j) + (M/2)*((-1)^m)*f*(1/(1+m+gammab(j)));

endendsemilogy(gammab_dB,Pbs,’b:’);hold on

8.

γΣ =1

N0

(∑Mi=1 aiγi

)2

∑Mi=1 a2

i

≤ 1N0

∑a2

i

∑γ2

i∑a2

i

=∑

γ2i

N0

Where the inequality above follows from Cauchy-Schwartz condition. Equality holds if ai = cγi wherec is a constant

9. (a) γi = 10 dB = 10, 1 ≤ i ≤ N

N = 1, γ = 10, M = 4

Pb = .2e−1.5 γ

(M−1) = .2e−15/3 = 0.0013.

(b) In MRC, γΣ = γ1 + γ2 + . . . + γN .So γΣ = 10N

Pb = .2e−1.5

γΣ(M−1) = .2e−5N ≤ 10−6

⇒ N ≥ 2.4412

So, take N = 3, Pb = 6.12 ×10−8 ≤ 10−6.

10. Denote N(x) = 1√2π

e−x2/2 , Q′(x) = −N(x)

P b =∫ ∞

0Q(

√2γ)dP (γ)

Q(∞) = 0, P (0) = 0d

dγQ(

√2γ) = −N(

√2γ)

√2

2√

γ= − 1√

2πe−γ 1

2√

γ

P b =∫ ∞

0

1√2π

e−γ 12√

γP (γ)dγ

P (γ) = 1− e−γ/γM∑

k=1

(γ/γ)k−1

(k − 1)!

1∫ ∞

0

1√2π

e−γ 12√

γdγ =

12

2∫ ∞

0

1√2π

e−γ 12√

γe−γ/γ

M∑

k=1

(γ/γ)k−1

(k − 1)!dγ =

M∑

k=1

1(k − 1)!

[1

2√

π

∫ ∞

0e−γ1+ 1

γ

γ−1/2

(γ

γ

)k−1

dγ

]

Denote A =(

1 +1γ

)−1/2

=M−1∑

m=0

1m!

12√

pi

∫ ∞

0e−γ/A2

γ−1/2

(γ

γ

)m

dγ

let γ/A2 = u

=M−1∑

m=0

1m!

12√

pi

∫ ∞

0e−u u−1/2

A

(uA2

γ

)m

A2du

=A

2+

M−1∑

m=1

(2m− 1

m

)A2m

22m

A

γm

P b =1−A

2−

M−1∑

m=1

(2m− 1

m

)A2m+1

22mγm

11.

DenoteN(x) =1√2π

e−x2/2 Q′(x) = [1− φ(x)]′ = −N(x)

Pb =∫ ∞

0Q(

√2γ)dP (γ) =

∫ ∞

0

1√2π

e−γ 1√2γ

P (γ)dγ

∫ ∞

0

1√2π

e−γ 1√2γ

dγ =1√π

Γ(

12

)=

12

(1)∫ ∞

0

1√2π

e−γ 1√2γ

e−2γ/γdγ =1

2√

1 + 2γ

(2)

∫ ∞

0

1√2π

e−γ 1√2γ

√πγ

γe−γ/γ

(1− 2Q

(√2γ

γ

))dγ =

12√

γ

1B√

Aγ(3)

where A = 1 +2γ

, B = 1 +1γ

overall P b =12

[1−

√1− 1

(1 + γ)2

]

12.P b P b(10dB) P b(20dB)

no diversity 12

[1−

√γb

1+γb

]0.0233 0.0025

two branch SC∫

Q(√

2γ)pγΣdγ 0.0030 3.67× 10−5

two branch SSC∫

Q(√

2γ)pγΣdγ 0.0057 1.186× 10−4

two branch EGC∫

Q(√

2γ)pγΣdγ 0.0021 2.45× 10−5

two branch MRC∫

Q(√

2γ)pγΣdγ 0.0016 0.84× 10−5

As the branch SNR increases the performance of all diversity combining schemes approaches the same.

MATLAB CODE:gammatv = [.01:.1:10];gammab = 100;gamma = [0:.01:50*gammab];for i = 1:length(gammatv)

gammat = gammatv(i);gamma1 = [0:.01:gammat];gamma2 = [gammat+.01:.01:50*gammab];tointeg1 = Q(sqrt(2*gamma1)).*((1/gammab)*(1-exp(-gammat/gammab)).*exp(-gamma1/gammab));tointeg2 = Q(sqrt(2*gamma2)).*((1/gammab)*(2-exp(-gammat/gammab)).*exp(-gamma2/gammab));anssum(i) = sum(tointeg1)*.01+sum(tointeg2)*.01;

end

13. gammab_dB = [10];gammab = 10.^(gammab_dB/10);Gamma=sqrt(gammab./(gammab+1));pb_mrc =(((1-Gamma)/2).^2).*(((1+Gamma)/2).^0+2*((1+Gamma)/2).^1);pb_egc = .5*(1-sqrt(1-(1./(1+gammab)).^2));

0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

γ

Pb(γ

)

MRCEGC

dB penalty ~ .5 dB

Figure 3: Problem 13

14. 10−3 = Pb = Q(√

2γb) ⇒ 4.75, γ = 10MRC Pout = 1− e−γ0/γ

∑Mk=1

(γ0/γ)k−1

(k−1)! = 0.0827ECG Pout = 1− e−2γR −√πγRe−γR(1− 2Q(

√2γR)) = 0.1041 > Pout,MRC

15. P b,MRC = 0.0016 < 0.0021P b,EGC

16. If each branch has γ = 10dB RayleighγΣ = overall recvd SNR = γ1+γ2

2 ∼ γe−γ/(γ/2)

(γ/2)2γ ≥ 0

BPSKP b =

∫ ∞

0Q(

√2γ)pγΣdγ = 0.0055

17. p(γ) where∫∞0 p(γ)e−xγdγ = 0.01γ√

x

we will use MGF approach

P b =1π

∫ π/2

0Π2

i=1Mγi

(− 1

sin2 φ

)dφ

=1π

∫ π/2

0(0.01γ sinφ)2dφ

=(0.01γ)2

4= 0.0025

18.

P b =(

1− π

2

)3 2∑

m=0

(l + m

m

)(1 + π

2

)m

; π =√

γ

1 + γ

Nakagami-2 fading

Mγ

(− 1

sin2 φ

)=

(1 +

γ

2 sin2 φ

)−2

P b =1π

∫ π/2

0

(Mγ

(− 1

sin2 φ

))3

dφ, γ = 101.5 = 5.12× 10−9

MATLAB CODE:

gammab = 10^(1.5);Gamma = sqrt(gammab./(gammab+1));

sumf = 0;

for m = 0:2f = factorial(2+m)/(factorial(2)*factorial(m));sumf = sumf+f*((1+Gamma)/2)^m;

endpb_rayleigh = ((1-Gamma)/2)^3*sumf;phi = [0.001:.001:pi/2];sumvec = (1+(gammab./(2*(sin(phi).^2)))).^(-6);pb_nakagami = (1/pi)*sum(sumvec)*.001;

19.

P b =1π

∫ π/2

0

(1 +

γ

2 sin2 φ

)−2 (1 +

γ

sin2 φ

)−1

dφ

gammab_dB = [5:.1:20];gammabvec = 10.^(gammab_dB/10);

for i = 1:length(gammabvec)gammab = gammabvec(i);phi = [0.001:.001:pi/2];sumvec = ((1+(gammab./(2*(sin(phi).^2)))).^(-2)).*((1+...

(gammab./(1*(sin(phi).^2)))).^(-1));pb_nakagami(i) = (1/pi)*sum(sumvec)*.001;

end

5 10 15 2010

−7

10−6

10−5

10−4

10−3

10−2

γavg

(dB)

Pb av

g

Figure 4: Problem 19

20.Pb =

23Q

(√2γb(3) sin

(π

8

))

α = 2/3, g = 3 sin2(π

8

)

Mγ

(− g

sin2 φ

)=

(1 +

gγ

sin2 φ

)−1

P b =α

π

∫ π/2

0

(1 +

gγ

sin2 φ

)−M

dφ

MATLAB CODE:M = [1 2 4 8];alpha = 2/3; g = 3*sin(pi/8)^2;

gammab_dB = [5:.1:20];gammabvec = 10.^(gammab_dB/10);

for k = 1:length(M)for i = 1:length(gammabvec)

gammab = gammabvec(i);phi = [0.001:.001:pi/2];sumvec = ((1+((g*gammab)./(1*(sin(phi).^2)))).^(-M(k)));pb_nakagami(k,i) = (alpha/pi)*sum(sumvec)*.001;

endend

5 10 15 2010

−15

10−10

10−5

100

γavg

(dB)

Pb av

g

Figure 5: Problem 20

21.

Q(z) =1π

∫ π/2

0exp

[− z2

sin2 φ

]dφ , z > 0

Q2(z) =1π

∫ π/4

0exp

[− z2

2 sin2 φ

]dφ , z > 0

Ps(γs) =4π

(1− 1√

M

)∫ π/2

0exp

[− gγs

sin2 φ

]dφ−

4π

(1− 1√

M

2)∫ π/4

0exp

[− gγs

sin2 φ

]dφ

P s =∫ ∞

0Ps(γΣ)pγΣ(γΣ)dγΣ

=4π

(1− 1√

M

)∫ π/2

0

∫ ∞

0exp

(gγΣ

sin2 φ

)pγΣ(γ)dγΣdφ−

4π

(1− 1√

M

)2 ∫ π/4

0

∫ ∞

0exp

(gγΣ

sin2 φ

)pγΣ(γ)dγΣdφ

But γΣ = γ1 + γ2 + . . . + γM = Σγi

=4π

(1− 1√

M

)∫ π/2

0ΠM

i=1Mγi

(− g

sin2 φ

)dφ−

4π

(1− 1√

M

)2 ∫ π/4

0ΠM

i=1Mγi

(− g

sin2 φ

)dφ

22. Rayleigh: Mγs(s) = (1− sγs)−1

Rician: Mγs(s) = 1+k1+k−sγs

exp(

ksγs1+k−sγs

)

MPSK

P s =∫ (M−1)π/M

0Mγs

(− g

sin2 φ

)dφ → no diversity

Three branch diversity

P s =1π

∫ (M−1)π/M

0

(1 +

gγ

sin2 φ

)−1 [(1 + k) sin2 φ

(1 + k) sin2 φ + gγs

exp(− kγsg

(1 + k) sin2 φ + gγs

)]2

dφ

g = sin2( π

16

)

= 0.1670

MQAM:Formula derived in previous problem with g = 1.5

16−1 = 1.515

P s = 0.0553

MATLAB CODE:

gammab_dB = 10;gammab = 10.^(gammab_dB/10);K = 2;

g = sin(pi/16)^2;phi = [0.001:.001:pi*(15/16)];

sumvec=((1+((g*gammab)./(sin(phi).^2))).^(-1)).*((((...(1+K)*sin(phi).^2)./((1+K)*sin(phi).^2+...g*gammab)).*exp(-(K*gammab*g)./((1+K)*sin(phi).^2+g*gammab))).^2);

pb_mrc_psk = (1/pi)*sum(sumvec)*.001;

g = 1.5/(16-1);phi1 = [0.001:.001:pi/2];phi2 = [0.001:.001:pi/4];

sumvec1=((1+((g*gammab)./(sin(phi1).^2))).^...(-1)).*(((((1+K)*sin(phi1).^2)./((1+K)*...sin(phi1).^2+g*gammab)).*exp(-(K*gammab*g)./((...1+K)*sin(phi1).^2+g*gammab))).^2);

sumvec2=((1+((g*gammab)./(sin(phi2).^2))).^(-1)).*((((...(1+K)*sin(phi2).^2)./((1+K)*sin(phi2).^2+...g*gammab)).*exp(-(K*gammab*g)./((1+K)*sin(phi2).^2+g*gammab))).^2);

pb_mrc_qam = (4/pi)*(1-(1/sqrt(16)))*sum(sumvec1)*.001 - ...(4/pi)*(1-(1/sqrt(16)))^2*sum(sumvec2)*.001;

5 10 15 2010

−9

10−8

10−7

10−6

10−5

10−4

10−3

10−2

10−1

100

Ps av

g

Figure 6: Problem 22

23. MATLAB CODE:

M = [1 2 4 8];alpha = 2/3;g = 1.5/(16-1);

gammab_dB = [5:.1:20];gammabvec = 10.^(gammab_dB/10);

for k = 1:length(M)for i = 1:length(gammabvec)

gammab = gammabvec(i);phi1 = [0.001:.001:pi/2];

phi2 = [0.001:.001:pi/4];sumvec1 = ((1+((g*gammab)./(1*(sin(phi1).^2)))).^(-M(k)));sumvec2 = ((1+((g*gammab)./(1*(sin(phi2).^2)))).^(-M(k)));pb_mrc_qam(k,i) = (4/pi)*(1-(1/sqrt(16)))*sum(sumvec1)*.001 - ...

(4/pi)*(1-(1/sqrt(16)))^2*sum(sumvec2)*.001;end

end

Chapter 10

1. (a)

(AAH)T = (AH)T .AT

= (AT )TAT

= AAH

∴ (AAH)H = AAH

For AAH , λ = λ, i.e. eigen-values are real

AAH = QΛQH

(b) XHAAHX = (XHA)(XHA)H = ‖XHA‖ ≥ 0∴ AAH is positive semidefinite.

(c) IM + AAH = IM + QΛQH = Q(I + Λ)QH

AH positive semidefinite ⇒ λi ≥ 0∀i∴ 1 + λi > 0∀i∴ IM + AAH positive definite

(d)

det[IM + AAH ] = det[IM + QΛQH ]= det[Q(IM + ΛM )QH ]= det[IM + ΛM ]

= ΠRank(A)i=1 (1 + λi)

det[IN + AHA] = det[IN + QΛQH ]= det[Q(IN + ΛN )QH ]= det[IN + ΛN ]

= ΠRank(A)i=1 (1 + λi)

∵ AAH and AHA have the same eigen-value∴ det[IM + AAH ] = det[IN + AHA]

2. H = UΣV T

U =

−0.4793 0.8685 −0.1298−0.5896 −0.4272 −0.6855−0.6508 −0.2513 0.7164

Σ =

1.7034 0 00 0.7152 00 0 0.1302

V =

−0.3458 0.6849 0.4263−0.5708 0.2191 0.0708−0.7116 −0.6109 0.0145−0.2198 0.3311 −0.9017

3. H = UΣV T

Let

U =

1 00 10 0

V =

1 00 10 0

Σ =

[1 00 2

]

∴ H =

1 0 00 2 00 0 0

4. Check the rank of each matrixrank(HI) = 3∴ multiplexing gain = 3rank(H2) = 4∴ multiplexing gain = 4

5.

C =RH∑

i=1

log2

(1 +

λiρ

Mt

)

Constraint∑

Vi = ρ∑

λi = constant

∴ ∂C

∂λi=

ρ

Mt ln 21

(1 + λiρMt

)− ρ

Mt ln 21

(1 + λiρMt

)= 0

⇒ λi = λj

∴ when all RH singular values are equal, this capacity is maximized.

6. (a) Any method to show H ≈ UΛV is acceptable. For example:

D =

.13 .08 .11

.05 .09 .14

.23 .13 .10

where : dij =

∣∣∣Hij−HHij

∣∣∣× 100

(b) precoding filter M = V −1

shaping filter F = U−1

F =

−.5195 −.3460 −.7813−.0251 −.9078 .4188−.8540 .2373 .4629

M =

−.2407 −.8894 .3887−.4727 −.2423 −.8473−.8478 .3876 .3622

Thus Y = F(H)MX + FN = U∗UΛVV ∗X + U∗N= ΛX + U∗N

(c) PiP = 1

γo− 1

γifor 1

γi> 1

γo, 0 else

γi = λi2P

NoB = 94.5 for i = 1, 6.86 for i = 2, .68 for i = 3Assume γ2 > γ0 > γ3 since γ3 = .68 is clearly too small for data transmission

∑ PiP = 1 ⇒ 2

γ0− 1

γ1− 1

γ2= 1 ⇒ γ0 = 1.73

P1P = .5676 P2

P = .4324C = B

[log2

(1 + γ1

P1P

)+ log2

(1 + γ2

P2P

)]= 775.9 kbps

(d) With equal weight beamforming, the beamforming vector is given by c = 1√(3)

[1 1 1]. The SNR

is then given by:

SNR =cHHHHc

N0B= (.78)(100) = 78. (1)

This gives a capacity of 630.35 kbps. The SNR achieved with beamforming is smaller than thebest channel in part (c). If we had chosen c to equal the eigenvector corresponding to the besteigenvalue, then the SNR with beamforming would be equal to the largest SNR in part(c). Thebeamforming SNR for the given c is greater than the two smallest eigenvalues in part(c) becausethe channel matrix has one large eigenvalue and two very small eigenvalues.

7. C = maxB log2 det[IMγ + HRXHH ]RX : Tγ(RX) = ρ If the channel is known to the transmitter, it will perform an SVD decomposition ofH as

H = UΣV

HRXHH = (UΣV )RX(UΣV )H

By Hadamard’s inequality we have that for A ∈ <n×n

det(A) ≤ Πni=1Aii

with equality iff A is diagonal.We choose RX to be diagonal, say = Ω then

det(IMR + HRXHH) = det(I + ΩΣ2)

∴ C = maxPi ρi≤ρ

Bσi log2(1 + λiρi)

where√

λi are the singular values.

8. The capacity of the channel is found by the decomposition of the channel into RH parallel channels,where RH is the rank of the channel matric H.

C = maxρi:P

i ρi≤ρ

∑

i

B log2(1 + λiρi)

where√

λi are the RH non-zero singular values of the channel matrix H and ρ is the SNR constraint.

γi = λiρ

Then the optimal power allocation is given as

Pi

P=

1γ0− 1

γiγi ≥ γ0

0 γi < γ0(2)

for some cut-off value γ0. The resulting capacity is given as

C =∑

i:γi≥γ0

B log2(γi/γ0)

For

H =

1 1 −1 11 1 −1 −11 1 1 11 1 1 −1

RH = 3, γ1 = 80, γ2 = 40, γ3 = 40. We first assume that γ0 is less than the minimum γi which is 40.

γ0 =3

1 +∑3

i=11γi

which gives γ0 = 2.8236 < mini γi, hence the assumption was correct.

C

B= 12.4732 bits/sec/Hz

For

H =

1 1 1 −11 1 −1 11 −1 1 11 −1 −1 −1

RH = 4, γ1 = 40, γ2 = 40, γ3 = 40, γ4 = 40. We first assume that γ0 is less than the minimum γi

which is 40.γ0 =

41 +

∑4i=1

1γi

which gives γ0 = 3.6780 < mini γi, hence the assumption was correct.

C

B= 13.7720 bits/sec/Hz

9. H =

h11 . . . h1Mt

. . . . .

. . . . .

. . . . .hMr1 . . . hMrMt

Mr×Mt

Denote G = HHT

limMt→∞

1Mt

Gii = limMt→∞

1Mt

[hi1 . . . hiMt ]

hi1

.

.

.

hiMt

= limMt→∞

1Mt

Mt∑

j=1

‖hij‖2

= Ej‖hij‖2

= σ2

= 1 ∀i

limMt→∞,i6=j

1Mt

Gij = limMt→∞

1Mt

[hi1 . . . hiMt ]

hj1

.

.

.

hjMt

= limMt→∞

1Mt

Mt∑

k=1

hikhjk

= Ekhikhjk

= Ek(hik)Ek(hjk)= 0 ∀i, j, i 6= j

∴ limM→∞

1M

HHT = IM

∴ limM→∞

B log2 det[IM +

ρ

MHHT

]= B log2 det [IM + ρIM ]

= B log2 [1 + ρ] det IM

= MB log2 [1 + ρ]

10. We find the capacity by randomly generating 103 channel instantiations and then averaging over it.We assume that distribution is uniform over the instantiations.

MATLAB CODE

clear;clc;Mt = 1;Mr = 1;rho_dB = [0:25];

rho = 10.^(rho_dB/10);for k = 1:length(rho)

for i = 1:100H = wgn(Mr,Mt,0,’dBW’,’complex’);[F, L, M] = svd(H);for j = 1:min(Mt,Mr)

sigma(j) = L(j,j);endsigma_used = sigma(1:rank(H));gamma = rho(k)*sigma_used;%% Now we do water filling\gammatemp = gamma;gammatemp1 = gammatemp;gamma0 = 1e3;while gamma0 > gammatemp1(length(gammatemp1));

gammatemp1 = gammatemp;gamma0 = length(gammatemp1)/(1+sum(1./gammatemp1));gammatemp = gammatemp(1:length(gammatemp)-1);

endC(i) = sum(log2(gammatemp1./gamma0));

endCergodic(k) = mean(C);

end

0 5 10 15 20 250

5

10

15

20

25

ρ (dB)

Cer

godi

c

Mt = M

r = 3

Mt = 2 M

r =3

Mt = M

r = 2

Mt = 2 M

r =1

Mt = M

r = 1

Figure 1: Problem 10

11. We find the capacity by randomly generating 104 channel instantiations and then averaging over it.We assume that distribution is uniform over the instantiations.

MATLAB CODE

clear;clc;Mt = 1;

Mr = 1;rho_dB = [0:30];

rho = 10.^(rho_dB/10);for k = 1:length(rho)

for i = 1:1000H = wgn(Mr,Mt,0,’dBW’,’complex’);[F, L, M] = svd(H);for j = 1:min(Mt,Mr)

sigma(j) = L(j,j);endsigma_used = sigma(1:rank(H));gamma = rho(k)*sigma_used;C(i) = sum(log2(1+gamma/Mt));

endCout(k) = mean(C);pout = sum(C<Cout(k))/length(C);while pout > .01

Cout(k) = Cout(k)-.1;pout = sum(C<Cout(k))/length(C);

endif Cout(k)<0;

Cout(k) = 0;end

end

0 5 10 15 20 25 300

5

10

15

20

25

ρ (dB)

Cou

tage

Mt = M

r = 3

Mt = 2 M

r =3

Mt = M

r = 2

Mt = 2 M

r =1

Mt = M

r = 1

Figure 2: Problem 11

12.

P (u?n < X) = P

(Mr∑

i=1

uini < X

)

=Mr∑

i=1

uiP (ni < X)

= P (ni < X)

∴ the statistics of u?n are the same as the statistics of each of these elements

13.

Σx = ‖u?Hvx‖= ‖u?Hv‖2‖x‖2

= vHHHu?Hu?Hv‖x‖2

= vHHHHv‖x‖2

= vHQHQv‖x‖2

≤ λmax‖x‖2

with equality when u, v are the principal left and right singular vectors of the channel matrix H

∴ SNRmax = λmax‖x‖2

N= λmaxρ

14.

H =

0.1 0.5 0.90.3 0.2 0.60.1 0.3 0.7

When both the transmitter and the receiver know the channel, for beamforming, u and v correspondto the principal singular vectors (or the singular vectors corresponding to the maximum singular valueof H). Notice that the singular values of H are the square root of the eigen values of HHH (WishartMatrix).

Using Matlab, we get that the maximum singular value of H is 1.4480 and the singular vectors corre-sponding to this value are

uopt =

−0.7101−0.4641−0.5294

and

vopt =

−0.1818−0.4190−0.8896

It is easy to check that uToptuopt = 1 and vT

optvopt = 1 and that

uTopt ∗H ∗ vopt = 1.4480

Since, during beamforming from eq. 10.17 in reader,

y = (uT Hv)x + uT n

and for a given transmit SNR of ρ, the received SNR is given as

SNRrcvd = ρ(uToptHvopt)2

since, uopt has norm 1, noise power is not increased. For, ρ = 1, SNR is simply (1.4480)2 = 2.0968.

When the channel is not known to the transmitter, it allocates equal power to all the antennas and sothe precoding vector (or the optimal weights) at the transmitter is given as

v2 =1√3

111

Defineh = Hv2

So, eq. 10.17 in the reader can be written as

y = (uT h)x + uT n

To maximize SNR we need to find a u2 of norm 1 such that (u′h) is maximized.

Using Matlab, we get that the maximum singular value of h is 1.2477 and the singular vector corre-sponding to this value is

u2 =

0.69410.50900.5090

It is easy to check that uT2 u2 = 1 and that

uT2 ∗ h = 1.2477

Alternatively, from MRC concept we know that:

u2 =hH

||h|| =

0.69410.50900.5090

where ||h|| is the L2 norm of h.

For a given transmit SNR of ρ, the received SNR is given as

SNRrcvd = ρ(uT2 Hv2)2

since, u2 has norm 1, noise power is not increased. For, ρ = 1, SNR is simply (1.2477)2 = 1.5567.

15. (a) ρ = 10 dB = 10

Pe = ρ−d

So to have Pe ≤ 10−3, we should have d ≥ 3, or at least d = 3. Solving the equation that relatesdiversity gain d to multiplexing gain r at high SNR’s we get

d = (Mr − r)(Mt − r)

⇒ 3 = (8− r)(4− r)

Solving for r we getr = 3.35 or 8.64

We have that r ≤ minMr,Mt, so r ≤ 4 and so r = 3.35. But we know that r has to be an integer.So, we take the nearest integer which is smaller than the calculated value of r, which gives us r=3 .

No credits for this part:

If we are allowed to assume that equations 10.23 and 10.24 hold at finite SNR’s too and we aregiven that we can use base 2 for logarithms, we can find the data rate as

R = r log2(ρ) = 9.96 bits/s/Hz

(b) With, r = 3, we can find d as

d = (Mr − r)(Mt − r) = (8− 3)(4− 3) = 5

For this value of d,Pe = ρ−d = 10−5

16. According to SVD of h√λ = 1.242

∴ C/B = log2(1 + λρ) = log2(1 + 1.2422.10) = 4.038bps/Hz

17.

H =[

.3 .5

.7 .2

]=

[ −.5946 .8041−.8041 .5946

] [.8713 0

0 .3328

] [ −.8507 .5757−.5757 −.8507

]

P= 10mWN0 = 10−9 W/HzB = 100 KHz

(a) When H is known both at the transmitter and at the receiver, the transmitter will use the optimalprecoding filter and the receiver will use the optimal shaping filter to decompose the MIMO chan-nel into 2 parallel channels. We can then do water-filling over the two parallel channels availableto get capacity.

Finding the γi’s

γ1 =λ2

1P

N0B= 75.92

γ2 =λ2

2P

N0B= 11.08

Finding γ0

Now, we have to find the cutoff value γ0. First assume that γ0 is less than both γ1 and γ2. Then(

1γ0− 1

γ1

)+

(1γ0− 1

γ2

)= 1

⇒ 2γ0

= 1 +1γ1

+1γ2

⇒ γ0 =1

1 + 1γ1

+ 1γ2

= 1.81

which is less than both γ1 and γ2 values so our assumption was correct.

Finding capacity

Now we can use the capacity expression as

C =2∑

i=1

B log2

(γi

γ0

)= 800 Kbps

(b) Total is

Essentially we have two parallel channels after the precoding filter and the shaping filter are usedat the transmitter and receiver respectively.

M(γ) = 1 + γKS(γ)

S

Finding K

K =−1.5

ln(5Pb)= .283

γK = γ0/K.

Finding γ0 or γK

We now find the cut-off γ0. First assume that γ0 < γ1, γ2. Notice that γ1 and γ2 have alreadybeen calculated in part (a) as γ1 = 75.92 and γ2 = 11.08.

(1γ0− 1

γ1K

)+

(1γ0− 1

γ2K

)= 1

⇒ 2γ0

= 1 +1

γ1K+

1γ2K

⇒ γ0 =1

1 + 1K

(1γ1

+ 1γ2

) = 1.4649

which is less than both γ1 and γ2 values, so our assumption was correct.

γK = γ0/K = 5.1742

Finding Rate R

Therefore the total rate, R is given as

R = B

[log2

(γ1

γK

)+ log2

(γ2

γK

)]

⇒ R = B4.97

This gives that R=497.36 Kbps (Obviously less than ergodic capacity).

(c) Since now we use beamforming to get diversity only, the transmitter and the receiver use theprincipal left and right eigen vectors of the Wishart Matrix HHH .

Once this is done the SNR at the combiner output is simply λmaxρ, where λmax is the maximumeigen value of the Wishart Matrix HHH and ρ is P

N0B

Finding γs

As given in the question, λmax is 0.7592 and ρ was calculated to be 100. So we get that γs = 75.92 .

Finding Pb

When using BPSK, γs = γb. Now we can use the expression for Pb for BPSK

Pb = Q(√

2γb

)= Q

(√2× 75.92

)=

0 Using the approx. given in the Ques.3.4× 10−35 Using Matlab

Credit is given for either value.

Finding Rate R

Since we are using BPSK and are given that B = 1/Tb, we get the rate using BPSK to beR=100 Kbps .

Comparing with previous part

Comparing with part (b), we can see that the rate R decreases by 397.36 Kbps and the Pb im-proves as Pb is now 3.4× 10−35 ∼ 0 whereas earlier it was 10−3.

(d) Therefore we see that we can tradeoff rate for robustness of the system. If we are willing to de-crease the rate at which we transmit, we can get more diversity advantage i.e. one strong channelwhich gives a much less value of Pb.

18. (a) clear;clc;Mt = 4;Mr = Mt;rho_dB = [0:20];

rho = 10.^(rho_dB/10);for k = 1:length(rho)

for i = 1:1000H = wgn(Mr,Mt,0,’dBW’,’complex’);[F, L, M] = svd(H);for j = 1:min(Mt,Mr)

sigma(j) = L(j,j);endsigma_used = sigma(1:rank(H));gamma = rho(k)*sigma_used;%% Now we do water filling\gammatemp = gamma;gammatemp1 = gammatemp;gamma0 = 1e3;while gamma0 > gammatemp1(length(gammatemp1));

gammatemp1 = gammatemp;gamma0 = length(gammatemp1)/(1+sum(1./gammatemp1));gammatemp = gammatemp(1:length(gammatemp)-1);

endC(i) = sum(log2(gammatemp1./gamma0));

endCergodic(k) = mean(C);

end

(b) clear;clc;Mt = 4;Mr = Mt;rho_dB = [0:20];

rho = 10.^(rho_dB/10);for k = 1:length(rho)

for i = 1:1000H = wgn(Mr,Mt,0,’dBW’,’complex’);[F, L, M] = svd(H);for j = 1:min(Mt,Mr)

sigma(j) = L(j,j);endsigma_used = sigma(1:rank(H));gamma = rho(k)*sigma_used;C(i) = sum(log2(1+gamma/Mt));

endCout(k) = mean(C);

end

19. using Matlab we get Cout = 7.8320

MATLAB CODE

clear;clc;Mt = 4;Mr = Mt;rho_dB = 10;rho = 10.^(rho_dB/10);

for k = 1:length(rho)for i = 1:1000

0 2 4 6 8 10 12 14 16 18 200

5

10

15

20

25

ρ (dB)

C

Cergodic

Mt = M

r = 1

Cout

Mt = M

r = 1

Cergodic

Mt = M

r = 4

Cout

Mt = M

r = 4

Figure 3: Problem 18

H = wgn(Mr,Mt,0,’dBW’,’complex’);[F, L, M] = svd(H);for j = 1:min(Mt,Mr)

sigma(j) = L(j,j);endsigma_used = sigma(1:rank(H));gamma = rho(k)*sigma_used;C(i) = sum(log2(1+gamma/Mt));

endCout(k) = mean(C);pout = sum(C<Cout(k))/length(C);while pout > .1

Cout(k) = Cout(k)-.01;pout = sum(C<Cout(k))/length(C);

endif Cout(k)<0;

Cout(k) = 0;end

end

20. As µ increases, the span of cdf becomes narrower and so capacity starts converging to a single number.

MATLAB CODE

clear;clc;

Mt = 8;Mr = Mt;rho_dB = 10;rho = 10.^(rho_dB/10);

for i = 1:1000H = wgn(Mr,Mt,0,’dBW’,’complex’);[F, L, M] = svd(H);for j = 1:min(Mt,Mr)

sigma(j) = L(j,j);endsigma_used = sigma(1:rank(H));gamma = rho*sigma_used;C(i) = sum(log2(1+gamma/Mt));

end[f,x] = ecdf(C);

6 7 8 9 10 11 12 13 14 15 16 17 180

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1Empirical CDF’s of Capacity w/o Tx CSI

x

FX(x

)

M = 4M = 6M = 8

Figure 4: Problem 20

Chapter 11

1. See Fig 1

fc = 100 MHz

fc+Bfc-B

2B = 100 KHz

Figure 1: Band of interest.

B = 50 KHz, fc = 100 MHz

Heq(f) =1

H(f)= f

Noise PSD = N0 W/Hz. Using this we get

Noise Power =∫ fc+B

fc−BN0|Heq(f)|2df (1)

= N0

∫ fc+B

fc−Bf2df (2)

= N0f3

3

](fc+B)

(fc−B)

(3)

=N0

3(fc + B)3 − (fc −B)3 (4)

= 1021N0 W (5)

Without the equalizer, the noise power will be 2BN0 = 105N0 W. As seen from the noise power values,there is tremendous noise enhancement and so the equalizer will not improve system performance.

2. (a) For the first channel:ISI power over a bit time = A2Tb/Tb = A2 For the 2nd channel:ISI power over a bit time = A2

Tb

∑∞n=1

∫ (n+1)Tb

nTbe−t/Tmdt = 2e−1/2A2

A

S(t)

Tm /2

1

h

Tm

1(t)

h2(t)

t

t

t

Figure 2: Problem 2a

(b) No ISI: pulse interval = 11/2µs = 5.5µs∴ Data rate = 1/5.5µs = 181.8KbpsIf baseband signal =100KHz: pulse width = 10µsData rate = 2/10µs + 10µs = 100Kbps

3. (a)

h(t) =

e−tτ t ≥ 0

0 o.w.(6)

τ = 6 µ sec

Heq(f) =1

H(f)

H(f) =∫ ∞

0e−

tτ e−j2πftdt (7)

=1

1τ + j2πf

(8)

Hence,

Heq(f) =1τ

+ j2πf

(b)

SNReq

SNRISI=

RB−B Sx(f)|H(f)|2|Heq(f)|2dfRB

−B N0|Heq(f)|2dfRB−B Sx(f)|H(f)|2df

2BN0

1

h

10us

1(t)

t

1

X

1us

(t)

t

h

12us

1(t)*X(t)

t

1

Figure 3: Problem 2b

Assume Sx (f) = S, −B ≤ f ≤ B ⇒2BS

N0

2Bτ2 + 8π2

3B3

SRB−B |H(f)|2df

2BN0

=

2B1

τ2 + 4π2

3B2

1.617× 10−6 = 0.9364 = −0.28 dB

(c)

h [n] = 1 + e−Ts

τ δ [n− 1] + e−2Ts

τ δ [n− 2] + ...

H(z) = 1 + e−Ts

τ z−1 + e−2Ts

τ z−2 + e−3Ts

τ z−3 + ...

=∞∑

n=0

(e−

Tsτ z−1

)n=

z

z − e−Tsτ

=1

1− e−Tsτ z−1

⇒ Heq(z) = 1H(z)+N0

. Now, we need to use some approximation to come up with the filter tap

coefficient values. If we assume N0 u 0 (the zero-forcing assumption), we get Heq(z) = 1−e−Tsτ z−1.

Thus, a two tap filter is sufficient. For Ts = 130 ms, we have a0=1, a1 = −0.0039 as the tap

coefficient values. Any other reasonable way is also accepted.

4. ωi = ci where ci is the inverse Z- transform of 1/F(z)Show that this choice of tap weights minimizes

∣∣∣∣1

F (z)− (ω−NzN + . . . + ωNz−N )

∣∣∣∣2

. . . (1)

at z = ejω

If F(z) is of length 2 and monic, say F (z) = 1− a1z then

1F (z)

= 1 + a1z−1 + a2

1z−2 + . . . where c1 = a1, c2 = a2

1, a1 < 1

It is easy to see that the coefficients become smaller and smaller. So if we had the opportunity tocancel any (2N+1) coefficients we will cancel the ones that are closest to z0. Hence we get that ωi = ci

minimizes (1). The result can be similarly proved for length of F(z) greate than 2 or non-monic.

5. (a) Heq(f) = 1H(f) for ZF equalizer

Heq(f) =

1 fc − 20MHz ≤ f < fc − 10MHz2 fc − 10MHz ≤ f < fc

0.5 fc ≤ f < fc + 10MHz4 fc + 10MHz ≤ f < fc + 20MHz0 o.w.

(9)

(b) S=10mW Signal powerN = N0[12 × 10MHz + 22 × 10MHz + 0.52 × 10MHz + 42 × 10MHz] = 0.2125mW∴ SNR = 47.0588 = 16.73dB

(c) Ts = 0.0125µsecPb ≤ 0.2e−1.5γ/M−1 or M ≤ 1 + 1.5SNR

−ln(5Pb)

for Pb = 10−3 M ≤ 14.3228 (M ≥ 4 thus using the formula is reasonable)R = log2 M

Ts= 307.2193Mbps

(d) We use M=4 non overlapping subchannels, each with B=10MHz bandwidth1: fc − 20MHz ≤ f < fc − 10MHz α1 = 12: fc − 10MHz ≤ f < fc α2 = 0.53: fc ≤ f < fc + 10MHz α3 = 24: fc + 10MHz ≤ f < fc + 20MHz α4 = 0.25Power optimization: γi = Pα2

iN0B for i = 1, 2, 3, 4

γ1 = 1000 γ2 = 250 γ3 = 4000 γ4 = 62.5for Pb = 10−3 K = 0.2831

Pi

P=

1γ0− 1

Kγiγi ≥ γ0/K

0 γi ≤ γ0/K(10)

We can see that all subchannels will be used and

P1 = 2.6523 P2 = 2.5464 P3 = 2.6788 P4 = 2.1225

andγ0 = 3.7207

thus R = 2B∑4

i=1 log(Kγi/γ0) = 419.9711Mbps

6. (a)

Ff(t) =

T |f | < 1/T0 o.w.

FZ(f) =1TS

∞∑n=−∞

F

(f +

n

Ts

)

= 1

∴ folded spectrum of f(t) is flat.

(b)

yk = y(kT + t0)

=∞∑

i=−∞Xif(kT + t0 − iT )

=N∑

i=−N

Xif ((k − i)T + t0)

= Xksinc(t0) +N+k∑

i=−N+k,i 6=k

Xif ((k − i)T + t0)

︸ ︷︷ ︸ISI

(c)

ISI =N+k∑

i=−N+k,i6=k

Xisin (π(k − i) + t0/Tπ)

π(k − i) + t0/Tπ

= sin(πt0/T )N∑

i=−N,i6=0

1πt0/T − πi

= sin(πt0/T )N∑

i=1

[ −1πt0/T − πi

+1

πt0/T − πi

]

=2π

sin(πt0/T )N∑

n=1

n

n2 − t20/T 2

Thus, ISI →∞ as N →∞7. gm(t) = g?(−t) = g(t) = sinc(t/TS), |t| < Ts

Noise whitening filter : 1G?

m(1/z?)

8. Jmin = 1−∑∞j=−∞ cjf−j

B(z) = C(z)F (z)

=F (z)F ?(z−1)

F (z)F ?(z−1) + N0

=X(z)

X(z) + N0

∴ b0 =1

2πj

∮B(z)

zdz

=1

2πj

∮X(z)

z[X(z) + N0]dz

=T

2π

∫ π/T

−π/T

X(ejωT )X(ejωT ) + N0

dω

∴ Jmin = 1− T

2π

∫ π/T

−π/T

X(ejωT )X(ejωT ) + N0

dω

=T

2π

∫ π/T

−π/T

N0

X(ejωT ) + N0dω

=T

2π

∫ π/T

−π/T

N0

T−1∑∞

n=−∞ |H(ω + 2πn/T )|2 + N0dω

= Ts

∫ −0.5Ts

−0.5Ts

N0

FΣ(f) + N0df

9.

VW J =(

∂J

∂w0, . . . ,

∂J

∂wN

)

J = wT Mvw? − 2<Vdw

?+ 1

∴ ∂J

∂w= 2Mvw

T − 2Vd

∂J

∂w= 0 ⇒ 2Mvw

T = 2Vd

⇒ wopt =(MT

v

)−1V H

d

10.

Jmin = Ts

∫ 0.5Ts

−0.5Ts

N0

FΣ(f) + N0df

∵ N0

FΣ(f) + N0≥ 0 ∴ Jmin ≥ 0

N0

FΣ(f) + N0≤ N0

N0= 1

∴ Jmin ≤ Ts

∫ 0.5Ts

−0.5Ts

1df = 1

∴ 0 ≤ Jmin ≤ 1

11.

FΣ(f) =1Ts

∞∑n=−∞

F

(f +

n

Ts

)

=1Ts

∞∑n=−∞

1 + 0.5e−j2π

f+ n

Ts

+ 0.3e

−j4πf+ n

Ts

MMSE equalizer :

Jmin = Ts

∫ 0.5/Ts

−0.5/Ts

N0

FΣ(f) + N0df

DF equalizer :

Jmin = exp

Ts

∫ 0.5/Ts

−0.5/Ts

ln[

N0

FΣ(f) + N0

]df

12. (a) G(f) is a sinc(), so theoretically infinite. But 2/T is also acceptable (Null to Null bandwidth)

(b) τ À T is more likely since T = 10−9secAs long as τ > Tb, get ISI and so, frequency selective fading

(c) Require Tb = Tm + T ⇒ R = 1Tm+T = 49997.5bps

(d) Heq(z) = 1F (z) for ZF equalizer

⇒ Heq(z) = 1d0+d1z−1+d2z−2

Long division yields the first 2 taps asw0 = 1/α0

w1 = −α1/α20

13. (a)

Hzf (f) =1

H(f)=

1 0 ≤ f ≤ 10KHz2 10KHz ≤ f ≤ 20KHz3 20KHz ≤ f ≤ 30KHz4 30KHz ≤ f ≤ 40KHz5 40KHz ≤ f ≤ 50KHz

(b) The noise spectrum at the output of the filter is given by N(f) = N0|Heq(f)|2, and the noisepower is given by the integral of N(f) from -50 kHz to 50 kHz:

N =∫ 50kHz

f=−50kHzN(f)df = 2N0

∫ 50kHz

f=0kHz|Heq(f)|2df

= 2N0(1 + 4 + 9 + 16 + 25)(10kHz)= 1.1mW

(c) The noise spectrum at the output of the filter is given by N(f) = N0(H(f)+α)2

, and the noise poweris given by the integral of N(f) from -50 kHz to 50 kHz. For α = .5 we get

N = 2N0(.44 + 1 + 1.44 + 1.78 + 2.04))(10kHz) = 0.134 mW

For α = 1 we get

N = 2N0(.25 + .44 + .56 + .64 + .69))(10kHz) = 0.0516 mW

(d) As α increases, the frequency response Heq(f) decreases for all f . Thus, the noise power decreases,but the signal power decreases as well. The factor α should be chosen to balance maximizing theSNR and minimizing distortion, which also depends on the spectrum of the input signal (whichis not given here).

(e) As α → ∞, the noise power goes to 0 because Heq(f) → 0 for all f . However, the signal poweralso goes to zero.

14. The equalizer must be retrained because the channel de-correlates. In fact it has to be retrained atleast every channel correlation time.Benefits of training

(a) Use detected data to adjust the equalizer coefficients. Can work without training information

(b) eliminate ISI.

15. N = 4

LMS-DFE: 2N+1 operations/iteration ⇒ 9 operations/iteration

RLS: 2.5(N)2 + 4.5N operations/iteration ⇒ 58 operations/iteration

Each iteration, one bit sent. The bit time is different for LMS-DFE/RLS, Tb (LMS-DFE)<Tb (RLS).But time to convergence is faster for RLS.

Case 1: fD = 100 Hz ⇒ (∆tc) ≡ 10 msec, must retrain every 5 msec.

LMS-DFE: R = 107

9 - 1000 bits5 msecs = 911 Kbps

RLS: R = 107

58 - 50 bits5 msec = 162 Kbps

Case2: fD = 1000 Hz ⇒ retrain every 0.5 msec

RLMS-DFE = 0 bps

RRLS = 72.4 Kbps

16. In the adaptive method, we start with some initial value of tap coefficients W0 and then use thesteepest descent method

Wk+1 = WK −∆GK . . . (1)

where ∆ is some small positive number and GK is the gradient of MSE = E|dk − ˆdk|2 is RWk − p

(Notice that 11.37 was a solution of gradient =0 , ∴ RW = p)

∴ GK = RWk − p = −E[εkYk?]

where Yk = [yk+L . . .yK−L]T and εk = ˆdk − dk

Approximately (1) can be rewritten as

Wk+1 = Wk + ∆εkYk?

Chapter 15

1. City has 10 macro-cellseach cell has 100 users∴ total number of users = 1000Cells are of size 1 sqkmmaximum distance traveled to traverse =

√2km

∴ time =√

230 = 169.7s

In the new setupnumber of cells = 105 microcellstotal number of users = 1000× 1002userstime =

√2×10

30×103 =1.69s∴ number of users increases by 10000 and handoff time reduces by 1/100

2. See Fig 1

(0,0)

(0,1)

v

u

(1,1)

(1,0)

60

Figure 1: Problem 2

D2 = (j20)2 + (i20)2 − 2(j20)(i20) cos(2π/3)

⇒ D = 2a√

i2 + j2 + ij =√

3R√

i2 + j2 + ij

3. diamond shaped cells, R= 100mDmin = 600mD = 2KRK = D

2R = 6002×100 = 3

N = K2 = 9

(a) number of cells per cluster = N = 9

(b) number of channels per cell = total number/N = 450/9 = 50

4. (a) R=1kmD=6kmN = Acluster

Acell=

√3D2/2

3√

3R2/2= 1

3(D/R)2 = 1362 = 12

number of cells per cluster = N = 12

D

SHADED CELLS USESAME FREQUENCY

Figure 2: Problem 3

(b) number of channels in each cell = 1200/12 = 100

(c)√

i2 + j2 + ij = 2√

3 ⇒ i = 2, j = 2

5. R=10mD=60mγI = 2γ0 = 4M = 4 for diamond shaped cells

SIRa =R−γI

MD−γ0=

R−2

4D−4= 32400

SIRb =R−4

4D−4= 324

SIRc =R−2

4D−2= 9

SIRa > SIRb > SIRc

6. γ = 2BPSKPb = 10−6 → Pb = Q(

√2γb) ⇒ γb = SIR0 = 4.7534

B = 50MHzeach user100KHz = Bs

SIR = 1M

(DR

)γ M=6 for hexagonal cellsa1 = 0.167a2 = 3

N > 1a2

(SIR0

a1

)2/γ⇒ N ≥ 9.4879 ∴ N = 10

Cu = 50

7. G = 100ξ = 1λ = 1.5With no sectorization

SIR =1

ξ3G(Nc − 1)(1 + λ)

= 4.7534

Nc = b26.2450c = 26

With sectorization, interference is reduced by a factor of 3

Nc = b76.7349c = 76

8. SINR = GPNc−1i=1 Xi+N

α = p(Xi = 1)N ∼ G(0.247Nc, 0.078Nc)Pout = p(SIR < SIR0)

(a) Pout = p

(GPNc−1

i=1 Xi+N< SIR0

)= p

(∑Nc−1i=1 Xi + N > G

SIR0

)

(b) X =∑Nc−1

i=1 Xi then X ∼ Bin(α,Nc − 1)p(x + N > G/SIR0) =

∑Nc−1N=0 p(n + N > G/SIR0|x = n)p(x = n)

p(x = n) =(

Nc − 1n

)αn(1− α)Nc−1−n

p(x + N > G/SIR0) =Nc−1∑

N=0

p(N > G/SIR0 − n|x = n)p(x = n)

=Nc−1∑

N=0

p

(N − 0.247Nc√

0.078Nc>

GSIR0

− n− 0.247Nc√0.078Nc

|x = n

)

=Nc−1∑

N=0

Q

(G

SIR0− n− 0.247Nc√

0.078Nc

)p(x = n)

(c) Nc = 35α = 0.5SIR0 = 5G = 150p = 0.0973

MATLAB

for i = 1:length(n)pn(i) = (factorial(Nc-1)./(factorial(n(i)).*factorial(Nc-1-n(i))))...

*alpha.^n(i)*(1-alpha).^(Nc-1-n(i));end

sump = 0; for i = 1:length(n)f = ((G/sir0)-n(i)-.247*Nc)/(sqrt(.078*Nc));sump = sump + .5*erfc((f)/sqrt(2))*pn(i);

end

(d) If x can be approximated as Gaussian thenx ∼ G((Nc − 1)α, (Nc − 1)α(1− α))x + N ∼ G(0.247Nc + (Nc − 1)α, 0.078Nc + (Nc − 1)α(1− α))

p(x + N > G/SIR0) = Q

(G

SIR0− (0.247Nc + (Nc − 1)α)√

0.078Nc + (Nc − 1)α(1− α)

)

(e) p= 0.0969 (very accurate approximation!)

9. defineγk =

gkPk

nk + ρ∑

k 6=j gkjpjk, j ∈ 1, . . . K

where,gk is channel power gain from user k to his base station nk is thermal noise power at user k’s basestationρ is interference reduction factor (ρ ∼ 1/G)gkj is channel power gain from jth interfering transmitter to user k’s base stationpk is user k’s Tx powerpj is user j’s Tx powerdefine a matrix F such that

Fkj =

0 k = jγ?

kgkjρgk

k 6= j

k, j ∈ 1, . . .K

u =(

γ?1n1

g1,γ?

2n2

g2, . . . ,

γ?KnK

gK

)

If Perron Ferbinius eigenvalue of F is less than 1 , then a power control policy exists. The optimalpower control policy is given to be P ? = (I − F )−1u

10. Matlab

D = 2:.01:10;R = 1;gamma = 2;Pdes = R^(-gamma);

for i = 1:length(D)Pint = 6*(.2*(D(i)-R)^(-gamma)+.2*(D(i)-R/2)^(-gamma)+.2*(D(i))^(-gamma)...

+.2*(D(i)+R/2)^(-gamma)+.2*(D(i)+R)^(-gamma));Pintbest = 6*((D(i)+R)^(-gamma));Pintworst = 6*((D(i)-R)^(-gamma));ASE(i) = log(1+Pdes/Pint)/(pi*(.5*D(i))^2);ASEbest(i) = log(1+Pdes/Pintbest)/(pi*(.5*D(i))^2);ASEworst(i) = log(1+Pdes/Pintworst)/(pi*(.5*D(i))^2);

end

2 3 4 5 6 7 8 9 100

0.05

0.1

0.15

0.2

0.25

0.3

0.35

D

AS

E(D

)

5 Case ASEBest Case ASEWorst Case ASE

Figure 3: Problem 10

11. Pt = 5WB = 100KHzN0 = 10−16W/Hz

Pr = PtK(

d0d

)3d0 = 1,K = 100

(a) D=2R2 users share the band availableEach user gets 50KHz

BASESTATION

USER

R=1Km

Figure 4: Problem 11a

(b) P b = 14γb

⇒ 10−3 = 14γb

⇒ γb = 250If D(n) = 2nR, number of users that share band = 2(2(n-1)+1)∴ each user gets 100KHz

2(2n−1) = Bu(n)interference is only from first tier

SIR(n) =PtK

(d0R

)3

N02 Bu(n) + 2

(PtK

(d0√

R2+D(n)2

)3) > 25

using Matlab , n = 4, SIR = 261.9253, D = 8R

Matlab

Pt = 5;R = 1000;sigma_2 = 1e-16;n = 1;

D = 2*n*R;

Bu = (100/(2*(2*n-1)))*1e3;K = 100;d0 = 1;Pdes = Pt*K*(d0/R)^3;Pint = 2*(Pt*K*(d0/sqrt(R^2+D^2))^3);Npower = sigma_2*Bu;

sir = Pdes/(Npower+Pint);while sir < 250

n = n+1;D = 2*n*R;Bu = (100/(2*(2*n-1)))*1e3;K = 100;d0 = 1;Pdes = Pt*K*(d0/R)^3;Pint = 2*(Pt*K*(d0/sqrt(R^2+D^2))^3);Npower = sigma_2*Bu;sir = Pdes/(Npower+Pint);

end

(c) ASE = (R1+R2)/B2km×2km

R1 = R2 = Bu(1) log(1 + SIR(1))Bu(1) = 50KHz, SIR(1) = 5.5899ASE = 0.6801bps/Hz/km2

12. B=100KHzN0 = 10−9W/HzK = 10P = 10mW per user

(a) 0 ≤ α ≤ 1 α is channel gain between cells.See MatlabIf α is large, interference can be decoded and subtracted easily so capacity grows with α as highSNR’s (beyond an α value) .For low SNR values (α less than a value) c decreases with increase in α as interference is increasedwhich cannot be easily decoded due to low SNR.

MATLAB CODE:

B = 100e3;sigma_2 = 1e-9;P = 10e-3;K = 10;ss = .001;

alpha = 0:.01:1;theta = 0:ss:1;for i = 1:length(alpha)

capvec = log2(1+(K*P*(1+2*alpha(i)*cos(2*pi*theta)).^2)/(sigma_2*B));C(i) = (1/K)*sum(capvec)*ss;

end

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.82

0.84

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

1.02

α

C(α

)/B

B = 100 KHz

Figure 5: Problem 12a

(b) C(K)↓ as K↑ because as the number of mobile per cell increases system resources get shared moreand so per user capacity C(K) has to fall.

MATLAB CODE:

BB = 100e3;sigma_2 = 1e-9;P = 10e-3;K = 1:.1:30;ss = .001;alpha = .5;theta = 0:ss:1;for i = 1:length(K)

capvec = log2(1+(K(i)*P*(1+2*alpha*cos(2*pi*theta)).^2)/(sigma_2*B));C(i) = (1/K(i))*sum(capvec)*ss;

end

0 5 10 15 20 25 300

1

2

3

4

5

6

K

C(K

)/B

B = 100 KHz

Figure 6: Problem 12b

(c) as transmit power P ↑, capacity C ↑ but gets saturated after a while as the system becomesinterference limited.

MATLAB CODE:

B = 100e3;sigma_2 = 1e-9;P = [0:.1:100]*1e-3;K = 10;ss = .001;alpha = .5;theta = 0:ss:1;for i = 1:length(P)

capvec = log2(1+(K*P(i)*(1+2*alpha*cos(2*pi*theta)).^2)/(sigma_2*B));C(i) = (1/K)*sum(capvec)*ss;

end

0 10 20 30 40 50 60 70 80 90 1000

0.2

0.4

0.6

0.8

1

1.2

1.4

P (in mW)

C(P

)/B

B = 100 KHz

Figure 7: Problem 12c